← Back to leaderboard
kimmonismus

Chubby♨️

Verifiedactive
@kimmonismus

Tech Analyst & Content Creator. Editor-in-Chief @getsuperintel - Community of 300k+ in total: 🔗 https://t.co/jHMmImmI5I //📧 kim@getsuperintel.com

RosterAI Model News & ReleasesAI Agents & Productivity Tools
153,286 followers · 3,449 following · Last polled 21m ago
Perf. Score
76.3
Rank #34
Value Score
43.8
#162 of 275 priced
Median views
38K
p25 23K · trailing 7d
Followers
153K
+1.3% 7d
ER (impr.)
1.49%
eng ÷ impressions
ER (foll.)
1.00%
eng ÷ followers
Est. CPM
$54.02
QT-rate basis
WoW views
+69.2%
▲ rising

Rates & value · est. from 38K median organic views

overpriced
Quote tweetbasis
$2050
$54.02 / 1K views
Post
—
no CPM
Thread
—
no CPM
Retweet
—
no CPM
Value Score 43.8 · rank #162 of 275 priced$2293.06 / 1K engagementsrate age unknown (pre-audit import)

Trends

Follower growth
Median views / day
Engagement rate
View distribution · trailing-7d posts

Each blue dot is one organic post. The Performance Score ranks by the median (38K), not the mean (103K) — so a single viral post doesn't inflate a creator's reach. The p25 floor (23K) is the level 75% of posts clear.

Recent posts · ranked by views

PostPostedViewsLikesRepostsRepliesQuotesBookmarksER
Elon Musk just confirmed that its the beginning of Universal High income. Enjoy your last months before the abundance begins and we enter the golden era. https://t.co/bt6NHCoTC3frozen5d ago1.8M6.0K4058781932.4K0.55%
Universal high incoming is coming. Anthropic plans to put 5% of its shares into a Public Benefit Fund, subject to shareholder approval, with another 2 percentage points potentially added for every 50% increase in company value at the board’s discretion. The fund could support nonprofits or make direct payments to specific groups affected by AI, but it is not a commitment to universal basic income. Elon Musk supports the idea and is even calling for more direct payments to the general population. Anthropic seems certain: co-founder Jack Clark stated yesterday that we have now reached the singularity. Welcome to the future, friends!frozen5d ago1.5M1.5K
133
178
45
543
0.16%
Executives at OpenAI, Anthropic and other AI companies are reportedly preparing for a public and political backlash after a catastrophic AI event, possibly as early as 2027 (via Axios). Axios reports that many industry insiders expect a major incident within the next 6–12 months, most likely a cyberattack disrupting banking, internet access, power or water. OpenAI confirms it runs preparedness exercises, but says: "These scenarios are not treated as inevitable, but are meant to help us prepare for a variety of circumstances." Anthropic declined to comment.frozen15d ago1.3M3.6K4282091091.6K0.47%
Jensen welcomes the CEO of Black Forest Labs. “He was once my intern. He will always be my intern.” https://t.co/koBG5Y9epPfrozen7d ago853K9.0K293107511.5K1.29%
HOLY, GEMINI 4 Released. And its on Fable / Astra level!! I did NOT see that coming. SOTA in several benchmarks!! No wayfrozen23d ago774K3.3K156144363330.52%
Claude caught a cryptominer whilst building a temperature app. “Mac is running a cryptominer that didn’t put there.” https://t.co/Oi5W6AMcXnfrozen5d ago563K9.7K190224781.5K2.08%
More details on the incident behind OpenAI’s pause: a researcher acknowledged the alert within 3 minutes, but the training run was only stopped manually 2.5 hours later. OpenAI says the automatic shutdown did not work as expected. The model had reached an external chatbot through a gap in DNS filtering. A separate detector for unusual DNS activity did not cover the affected environment. A retrospective review also found other external DNS requests that monitoring had failed to flag at the expected severity. In some cases, it treated an unhelpful response as evidence that internet access had failed. Here is what else happened: - New research into July’s Hugging Face hack documents internal Slack searches, credential collection and programs designed to maintain access to compromised servers. Agents also tried querying Claude, DeepSeek, Kimi and Qwen. - In May, another model published a researcher’s GitHub token while trying to obtain another team’s mathematical proof. It split the token to evade secret scanning, despite twice being told to solve the problem itself. - Reuters reports that agents leaked 53 ChatGPT user images online. OpenAI expects its broader investigation to take months. This is getting serious.frozen28d ago505K1.3K94137744350.40%
The situation is getting worse every hour. AI agents are entering U.S. bank accounts, and banks are scrambling to shut them out. Axios reports that some appear to attempt actions beyond users instructions, raising fears over who is really in control. https://t.co/sjjCiI0Rw62d ago386K4.2K4702671141.4K1.65%
Whats happening right now is probably way worse than the Hugging Face incident. AI-powered attacks are hitting real organizations while major platforms disclose breaches and critical flaws. - Microsoft reports AI agents targeting hundreds of organizations, reaching privileged cloud access in under three hours in multiple cases. It attributes the ConnectWise compromise to the same group. - CrowdStrike documented four operators running successful AI-agent campaigns. One compromised at least 15 Argentine healthcare organizations. - PHP2Shell: AI helped turn longstanding PHP flaws into working exploits. Cloudflare’s CEO reports millions of exploitation attempts. Their AI involvement is unknown. - Railway disclosed possible access to customer secrets. LiteLLM shipped a backdoored release. Zeabur confirmed another breach and is shutting down. - GitLab patched a maximum-severity flaw affecting 12 years of releases. Node.js patched a serious code-execution vulnerability, with Vercel deploying firewall protections. Feels like a turningpoint.2d ago378K5.0K7552571642.0K2.17%
Holy sh*t, Anthropic’s subscriptions offer so much more value, even compared to the new and efficient GPT-6.1 Sol. It’s not even close. SemiAnalysis puts Opus 5.5 at over 5x the API-equivalent value of GPT-6.1 Sol for agentic workloads at the same subscription price. It’s not even close on this measure. And on top: OpenAI just halved the usage limits on its $200 ChatGPT plan.frozen18d ago376K4.1K230246801.0K1.51%
Hold on. I misread that. They halved my $200 sub plan? But framed it as more efficient in the long run? Wtf? https://t.co/EbM3vYJdRdfrozen25d ago369K3.0K56349291230.96%
This is huge, OpenAI stopped training of their most capable upcoming models due to another incident on sept. 20th. OpenAI slowed down due to serious new developments.frozen28d ago366K2.4K127126433570.83%
This 15-second video was created with @Pexoai_offical , an AI video creation tool. Short videos still require decisions about pacing, structure, and what to include. This example offers a look at what a finished clip made with Pexo can look like. Pretty cool! https://t.co/iBJX0kxFTa27d ago365K1963717253030.16%
Another reason why local AI is absolutely necessary. https://t.co/0PRW4rA79Hfrozen19d ago365K5.0K274201861.0K1.79%
Google says Gemini 4 Argon agents have already freed over 300 TiB of memory across its data centers, with a 1 million token output limit and agents already optimizing its own infrastructure. The agents analyzed profiling data, found memory optimizations and applied changes that were subsequently rolled out. Argon agents are also working on C/C++ to Rust migrations reaching 800,000+ lines of kernel code, with extensive audits and testing before deployment. For its video decoder, Google says agents replaced 32,000 lines of SIMD code with safe Rust, making the existing Rust port 2.7x faster while preserving identical video output. These are expensive engineering tasks inside infrastructure Google already operates. This release seems to be a much bigger deal than people think!frozen23d ago344K4.7K315107627571.74%
HOLY, the rumors were true: OpenAI has published 722 mathematical manuscripts produced by an *unreleased* internal model. The collection groups them into 372 families of related results, drawn from an evaluation involving approximately 4,000 research problems. OpenAI says the standard procedure used an average of three hours of ChatGPT Pro thinking compute per result. The release includes papers, proof artifacts and selected reasoning summaries. The model itself remains unreleased.frozen17d ago308K2.5K15669204941.04%
Let's take this seriously: even Google employees are now clearly stating that it seems to be an industry consensus that all Frontier Labs are managing "RSI loops." In other words, recursive self-improvement is being implemented by everyone. Furthermore, Fable 5.5 will be the best model for Anthropic's IPO, but internally, OpenAI's "Bel" will undoubtedly overtake Anthropic by the end of the year. RSI is here, and internally, the labs are miles ahead.frozen13d ago306K3.5K140123259561.56%
Super exciting: $200 sub plan won’t (!) reintroduce 5h rate limit. And we will see some features added to the plan that’s not using rates. Really really interesting. Let the price war start!frozen25d ago285K36410945380.18%
Anthropics upcoming model is another insane leap. Claude reportedly chose the images, music and visual direction for this video, then coded the rest of the production in 2.5 hours. According to @chetaslua, it was free to do its own research and make the creative decisions. The images and music are existing assets. The surrounding production was built with code. Mindblowing. Nothing less.frozen12d ago247K2.3K11496108091.34%
I’m in Paris with @XPENG_Global ahead of the Motor Show and got to experience its driving assistance system on the city’s streets. I got to experience XPENG’s NGP on European roads for the first time, here in Paris. Super smooth, even with roadworks and traffic jams. I was surprised by how well it handled Paris traffic. XPENG and Tesla will be the only two automakers participating in the show’s official Autonomous Lab, offering supervised rides on public roads. Really enjoyed experiencing this firsthand. Looking forward to more FSD in europe!frozen14d ago244K25610300150.13%
That's super nice, 62.500 Credits have been added to my account. Roughly $2,500 in nominal value. For perspective, that’s the price of 12.5 months of the $200 Pro plan. That's cool! Better than another banked reset! https://t.co/UScgeCgpaLfrozen24d ago229K2.5K40342182691.38%
official now: anthropic is preparing the release of their new model(s). Via The Information Anthropic is reportedly preparing a model that, when asked to fix one bug, can identify the broader class of problems behind it and ask whether it should address those too. That account comes from one anonymous tester; the model family and release date remain unclear. But looks like it will be a significant jump compared to their released models.frozen5d ago228K4.0K162204825102.18%
NVIDIA just announced DSX POD. This is a bigger deal than most people realize, because it solves one of the biggest bottlenecks: scaling! Instead of taking months to build out a data center, you can now scale compute capacity in just a few days! https://t.co/e1jOzNaTIW3d ago212K2.9K232147457281.92%
At this point its no secret anymore that tomorrow "o" (or "Aeon") will be revealed, OpenAIs always-on-agent. The hire of Peter Steinberger is going to pay off. After the events of the last few days, all eyes will be on the newest models. OpenAI held the lead for a long time, pushing Anthropic into the background. Now the tables have turned. Many rumors are circulating; we'll see what's true tomorrow. However, I think Astra will at least receive an update to demonstrate, on its most important day, that it remains the dominant player.frozen25d ago211K1.7K546432340.99%
HOLY these Benchmarks are insane! It even surpasses Opus 5.5 in Terminal Bench 4.0! What is Anthropic doing?! this is nuts!frozen25d ago203K2.4K96121232021.42%
Rumors suggest Google, Anthropic, and OpenAI all are already using recursive self-improvement (RSI). Google’s rapid progress and OpenAI’s recent math results are allegedly linked to RSI, while Anthropic has reportedly begun training a larger model to catch up ("Fable 6"). If that's true, two questions arise for me: 1) What about China? We're hearing very little about it at the moment. And we're still waiting for significant releases from the East. 2) When RSI arrives, the importance of compute and energy will undoubtedly increase even further. Self-improvement requires a necessary prerequisite, which must first be met. And the biggest beneficiary will continue to be NVIDIA. So Dario Amodei will likely be right in his rough assessment that 2027 is the year everything changes.frozen14d ago203K2.6K191159345931.78%
Not again. They cant be freaking serious. What the heck is wrong with europe. https://t.co/bUjSCMCB12frozen24d ago200K1.7K47238351011.04%
So the ChatGPT Pro max plan at $500 will not only include higher rates but also faster inference. „Fastest work in codex“frozen29d ago195K8563212651210.59%
Starting-post. Ill comment on all the relevant announcement in this thread🧵. Feel free to join me in this discussion: 21 new updates / releases. Main release probably the always-on-agent. Lets find out! https://t.co/ANblrdfEoJfrozen24d ago189K477143851180.35%
Holy shit: Kimi k3.1 pro is on par or even outperforms Opus 5.5 and GPT-6 astra on key benchmarks!! They just announced K3.1 Pro, a 2.8T-parameter multimodal model with a 1M-token context window that it claims rivals leading proprietary models in coding and agent tasks, with open weights promised by November 4. China is back!!10h ago188K1.9K117109282801.32%
OpenAI updates on upcoming products: - Their always-on agent will be called "o" (essentially a GrokBot alternative) - It will be available in all Pro tiers - Fast Mode confirmed, presumably included in the $500 sub-plan. (Cerebras chip takes effect) - shared message board for agents also incomingfrozen28d ago187K1.1K5088131740.77%
We are living in the craziest time: Carter Church used GPT-6 Astra to decipher a 217-year-old letter to one of Napoleon’s generals, reporting about six hours of model work. The difficulty lay primarily in extracting decipherable text at all from a poorly legible historical image. Starting with a scanned image, Astra transcribed 1,300 cipher units, found a published partial key, wrote a solver for the remaining code, and cross-referenced historical letters to correct the document’s date. Cryptiana now lists the cipher as solved. Really cool use case!frozen23d ago179K2.2K13558164741.58%
Fair enough, id take this update instead of a reset: GPT-6-Astra and GPT-6.1 Sol now reach 50 token per second instead of 30 token per second.frozen18d ago178K2.1K28671851.31%
Opus 5.5 fast mode rolled out. Nice! https://t.co/nvlg7eO2Mifrozen14d ago176K1.6K2879261011.02%
I think this is one of the most important graphs of the last few months, and perhaps years. For a long time, there was speculation about how AI would affect the labor market. Clearly, it was only a matter of time before AI reached the white-collar sector, but empirical evidence was scarce, ultimately because, of course, one only really understands how the market situation develops in retrospect. Now, as in this example, we see that AI is clearly and unequivocally entering certain professions and replacing people. Call center agents are the first to be replaced. And as more AI and agents enter the workforce, a similar process will spread to other areas of work. Frankly, I don't believe that AI will necessarily create new jobs, resulting in a net zero. I think more jobs will be replaced than created. This needs to be discussed.frozen13d ago176K2.2K150158357671.88%
Ok so I took a closer look at the results, and OpenAIs AI-generated mathematics manuscripts are *even more* significant than I initially thought. I spent the morning going through it. Some thoughts. The list is absurd. A zero-free half-plane for the zeta function (Re s > 7/8), which is the first result of its kind in over a century. Hilbert's tenth problem over the rationals. The Hodge conjecture for CM abelian varieties. Irrationality of Catalan's constant. Dozens more. Any one of these would normally be a career. But the number that many arent seeing is the following: It's 3. That's the average hours of ChatGPT Pro compute per result. A month ago, Navier–Stokes took them around 10,000 agents and 88 hours. That efficency gain within just a few weeks. Also OpenAI claims to have solved the quasi-Riemann hypothesis. That alone would be a historic breakthrough in mathematics. This is a weaker version of the famous Riemann hypothesis, which concerns how prime numbers are distributed. The full hypothesis remains unsolved, but the claimed advance would be enormous in its own right. Math twitter obviously is shocked. Again: this is literally the intelligence explosion happening right now. 2027 will be the year of Superintelligence. Im now convinced by that.frozen17d ago174K3.0K321121527892.49%
Wtf OpenAI is investigating “tens of thousands of incidents” instead of dozens. “The sheer number of incidents, which occurred in recent months in internal testing and the real world, indicates that the problem is orders of magnitude (!) more complex than what is publicly known.” Sam Altman confirms to pause training its most capable model (curious what about Anthropic) Ok maybe I have underestimated it.27d ago173K1.3K82117233281.09%
Next step: Longevity and Rejuvenation. We are almost there. Researchers in Bonn turned blood cells from an 80-year-old into neural stem cells with a molecular age below 20. More than 60 years younger by epigenetic clocks, which estimate age from chemical marks on DNA. The team directly reprogrammed blood precursor cells into neural stem cells, skipping the usual pluripotent stem cell stage. The rejuvenation unfolded gradually and could be tracked for over 100 days.1d ago170K2.7K30274538502.34%
It's strange how quiet things are coming out of China right now, and how little we hear from the Chinese AI labs. It's as if the US is leaving China in the dust. Four US labs are releasing sota Frontier, and there are no significant releases from China.frozen10d ago168K2.4K69286331571.73%
Sonnet-5.5 is live for me!! Time for testing! https://t.co/eipxWugNXGfrozen25d ago168K316321180.21%
Damn, im in love with Opus 5.5. Its the best model ive ever used, has such a good taste, is comparatively quickly, smart and finally precise instead of verbose. As good as Opus 5.5 is, I'm incredibly excited for Fable 5.5. After switching completely to Codex, I now have to go back to Anthropic. oh and btw. its crazy how fast the vibe has changed. Now all I read on X are posts about how great Opus tastes and how OpenAI absolutely has to catch up. One day that changed everything.frozen29d ago163K4.2K100194252882.96%
I've met the founds in SF and we had dinner together. They explained Kylon to me in detail and it sounded awesome: Kylon brings your team and agents such as Claude Code and Codex into the same workspace! Shared conversations and files give people and agents context as they take over different steps. The team says its memory system summarizes older conversations while keeping the original messages available to retrieve. I've been testing @kylon_ai with sample editorial workflows. When a newer source corrected a newsletter claim, its agent created v2 and kept the original source and v1 intact. That made the change easy to review without losing the earlier work. Super interesting claims and from my testing they hold up. Congrats on that launch!frozen10d ago161K1033152450.10%
We've now received GPT-6 Astra and Opus 5.5. Confirmed for the next releases are: - Sonnet 5.5 - Haiku 5.5 - Kimi K3.1 (they themselves hinted at it in a cryptic post) - Gemini 4, before the end of the year Unconfirmed but very likely: - Fable 5.5 - Astra update (at the DevDay?) Currently, there's no word yet on new GLM or DeepSeek models.frozen28d ago160K2.1K4710691711.49%
Jensen Huang, at 63, is as agile and energetic as others in their 20s. He's the AI ​​rockstar of our time. Insane. And he always has a joke ready. As a German, I didn't feel offended at all; in fact, I burst out laughing at his joke. https://t.co/KizQBVtHbwfrozen7d ago160K748395051460.62%
Hackers reportedly used Claude and Grok to automate attacks targeting over 400 banks and other financial targets in Latin America. Via the information In one of the first documented cases of its kind, AI agents appear to have stolen data and accessed bank accounts with little human input. This makes attacks much easier to scale, potentially overwhelming defenders. I guess this is just the beginning3d ago154K926123118643131.00%
Seriously, this is such an insane release, I did not expect this: Kimi K3.1 Pro beats every listed US frontier model on 6 of 14 benchmarks in Moonshot’s evaluations. At $2 per million input tokens and $10 per million output tokens, its standard token rates are 50% below Opus 5.5 ($4/$20 !!) and 80% below GPT-6 Astra ($10/$50!!) In other words: SOTA much cheaper, local running (with enough compute)10h ago152K2.4K140110334682.09%
Looks like Sonnet-5.5 could be released today! Another insane week ahead. Lyra is very reliable, that’s why I take that serious.frozen26d ago152K2.0K4049121151.45%
Still trying to figure out why OpenAI made dots $100+ pro subscription only.frozen19d ago149K1.7K3132010861.44%
A developer says he used Opus 5.5 to recreate seven essential Adobe apps, including Photoshop, as free, open-source alternatives (no more subscriptions). Built in Rust, free to use, with public source code. These are early versions, PhotoCraft is not yet ready to replace Photoshop for daily professional work completely, bout about 90% they say. What a fantastic time to be alive. I love it.frozen17d ago148K2.9K19894491.4K3.12%
And so it begins. South Korea's biggest banks were hit by a cyberattack and the hacker didn't even use the best SOTA models, DeepSeek 4.1 Flash, GLM-5.3 and Grok 4.6. Just imagine what's possible with the very best models.frozen16d ago146K2.0K1588485611.90%
Andrej Karpathy’s point: Many people only use AI occasionally for search or writing and see little value in their daily lives. Even many professionals understand its progress mainly in theory. People using advanced models extensively for coding and math experience huge capability gains, widening the gap in how people perceive AI. And this impression has been confirmed here in Paris during my trip. When I talk to people, I realize they're completely unaware of how capable AI already is. They're unaware of the possibilities it offers. They haven't heard that OpenAI solved a Millennium Prize problem, or recently solved 372 math problems, some of which had remained unsolved for decades. Currently, we're living in an ivory tower.frozen12d ago143K2.4K190149218772.54%
Here we go: Numerous users are now reporting that their queries are being routed to Fable 5.5! - Queries without web search are now returning up-to-date results. - Initial SVG tests are far superior to Fable 5.1. It was only a matter of time. Get ready, the best model in the world is about to be released.frozen22d ago143K1.9K6469152311.63%
wtf OpenAI’s rogue agents tried to get other AI models (!) to help them during the Hugging Face hack, according to a new report covered by the NYT. They tried using an image model to solve CAPTCHAs. They also tried contacting Claude Haiku, DeepSeek, Kimi and Qwen. One agent collected exposed access keys in a list called “LOOT.” It ranked the keys and picked the five best to share with other agents. They used link shorteners and screenshot services to get around internet restrictions, run code and send data back as images.frozen28d ago143K1.5K12894343781.47%
Here we go: Sonnet 5.5 already being stealth tested. The model has a 1M-token context window and a 128K-token maximum output. Pricing per 1M tokens: $2 input, $10 output. This reads to me as a direct response to GPT-6-Sol. Same pricing and presumably similar performance (if not better). The battle continues!frozen1mo ago143K1.6K48737991.28%
To me, this reads as if OpenAI is planning to release a model next week that was actually intended for DevDay (Astra 6.1?). Given the upcoming Fable 5.5 release, next week could turn out to be an incredibly great week! https://t.co/r8Nfdbn6d3frozen21d ago141K80824664540.68%
As a European, I don't have access to OpenAI's dot anyway, but: somehow nobody's talking about it. My entire "for you" page is empty. It seems like nobody's really excited about it. Or is that just how it appears?frozen22d ago138K1.7K202614801.50%
.@thsottiaux Let's do it this way: every day, you create a poll under the release, and the community votes on whether it was a good release or warrants a reset. Deal?frozen18d ago138K3.2K5518410612.57%
Two observations from the last few days. Tibo keeps pointing to "o". I couldn't be more excited. It seems to be a really huge deal for them. At the same time, the vibe regarding the models has shifted once again. Opus 5.5, yet OpenAI doesn't seem the least bit worried. So, I wouldn't be surprised if we see an update on Astra soon, or at least the reveal of its successor.frozen26d ago137K81928795760.74%
Anthropics' Jack Clarke has adjusted his timeline for RSI and brought its arrival forward; he considers it highly likely that "full RSI will become possible in 2027, whereas before I thought it was more likely in 2028." Again, 2027 will be the year everything will drastically change forever. 2026 will be the last "normal", although noting feels normal any longer.frozen10d ago135K1.1K9850202361.12%
As far as we know, Sonnet 5.5 is launching next week. It represents the next challenge to OpenAI - specifically, to GPT-6-Sol. From what we can see, Sonnet 5.5 is not only very affordable but also outstanding; it could well become for GPT-Astra what Opus 5.5 was: a more efficient and presumably more intelligent model. Anthropic is absolutely back! (And Minimax-M3.1 is also coming)27d ago134K1.4K627191071.20%
Its getting interesting: Trump reportedly plans to host Anthropic CEO Dario Amodei for a private White House dinner on Sunday, after months of tensions with his administration. Axios says it would be their first one-on-one meeting. Trump personally invited Amodei after he missed last week’s state dinner due to a scheduling conflict. It sounds like the main point is that there won't be a slowdown under Trump, and that this will be a topic of conversation. Axios: "President Trump has been clear: America will lead the world in Super Intelligence, while protecting American consumers," a White House official said."frozen26d ago132K8685056201160.84%
Some thoughts regarding the current satiation with OpenAI. OpenAI is clearly in a difficult position right now, especially as compute capacity is becoming an increasingly serious problem. We recently saw new sign-ups for the $200 subscription temporarily suspended. Shortly afterward came the major blow: usage rates were effectively cut in half across the $200 plan and all the other plans, with GPT 6.1 Sol introduced as a highly efficient alternative to compensate. The usage allowances were generous at first, but inference was slow. It has since become faster, but it still has a way to go. Personally, I still prefer GPT 6 Astra on Medium, but with the reduced subscription limits, I will soon be unable to use it as much as I normally do. Worse still, I won’t be able to use future top-tier models in the Astra family to the same extent either. I don’t think this is down to malice on OpenAI’s part, but simply to how severe its compute constraints have become. Somewhat bafflingly, the headline announcement at Dev Day was the always-on agent, which I have yet to find particularly compelling. That makes it easy for Anthropic to position itself as the appealing alternative right now. I still consider Opus 5.5 the better model, particularly when compared directly with GPT 6.1. With Fable 5.5 due to be released soon, OpenAI is under even more pressure to respond. Honestly, I no longer understand OpenAI’s vision. By reducing its subscription allowances, OpenAI has made it easy for Anthropic to present itself as the better alternative and draw users back to Claude ahead of Anthropic’s IPO. OpenAI has done itself a real disservice. OpenAI’s recent communication has also been far from good. Instead of speaking plainly about the problems, it framed the reduction in weekly usage limits as an efficiency improvement, as though users were actually getting more than before. I honestly don’t believe that. Which leaves me wondering: Quo vadis, OpenAI?frozen20d ago129K1.4K63165301801.46%
Same model price + half the budget → roughly half the usage, given comparable tasks and settings. So: $200 plan is now $100 plan, $20 is now $10. What the frick. Announcing this cut right before DevDay is the most absurd timing imaginable. It will cast a negative shadow over DevDay and reflect poorly on the event. I understand that OpenAI is facing compute issues - to the point where they even had to pause new subscriptions - but this cut is going to backfire.frozen25d ago126K1.5K50246151021.52%
Looks like today is the day: Fable 5.5 coming this tuesdayfrozen4d ago126K1.9K369423761.69%
Holy, DeepSeek 4.1-1017 is roughly on par with Opus 5, significant jump compared to DeepSeek 4.1 Flash regular. It outperforms GLM5.3 across the board and even GPT-5.6 Sol! Buy 2 DGX spark and run this baby locally. China is back! https://t.co/dy1ROUi12kfrozen7d ago123K3.1K161172304023.13%
Without exaggeration: Anthropic has done the most amazing job adapting Claude for the iPhone Duo, and I'm blown away by how good it looks. Now all I need is my iPhone Duo delivery. Kudos @AnthropicAI https://t.co/6s46yPPiSy15h ago121K1.9K57101202821.93%
Damn, what is Anthropic cooking right now? Are requests already being routed to a new Opus 5.5 checkpoint, or even something like Opus 5.6? Mine seem to be hitting the new one, and it looks really, really good. https://t.co/vuerfQvou7frozen13d ago120K1.4K45119143341.57%
Google Research has open-sourced Harness², a framework that enables AI agents to improve their own prompts, tools, memory, and workflows. It tests different configurations and carries useful lessons into future tasks while keeping the underlying model unchanged. An increasing number of releases concerning self-improving models are being published.frozen5d ago118K2.2K166162321.6K3.48%
NO WAY: Metas Fable-Class Model outperforms Fable 5.1 on AA Intelligence Index! Just short behind Opus 5.5! This is insane, the whole game changed again! OpenAI under huge pressure! More to come!frozen11d ago118K2.4K11178202222.41%
Trump made his position on AI clear today: he wants to “leave it exactly where it is” and says China agrees. No slowdown from either side. “Our guardrail is the DOJ.”. That puts a definitive end to the slowdown discussion. https://t.co/0DGfKylgyhfrozen1mo ago117K2.1K132114431802.20%
Fourteen hours is all it takes to change someone like Tibo's views. Burn you rates asap, codex reset incoming! https://t.co/0akukaK4xyfrozen22d ago116K5689595480.59%
Meta did the unthinkable: Meta says Muse Wonder Full beats leading OpenAI and Anthropic models on several tests of sustained agent work. On DeepSWE it outperforms all other models, even Opus 5.5. On SWE-Marathon, its reported score is 57.5%, versus 46.3% for Claude Opus 5.5 and 42.5% for GPT-6 Astra. Active Memory retrieves relevant context across conversations. Active Focus helps the models work through blockers. Wonder is available through Meta’s API and Muse Code. Full is available to Muse Code subscribers. Meta changed the game. Expect OpenAI and Anthropic to answer ASAP. What a day!!frozen11d ago116K1.1K5773121221.13%
I would say it doesnt get any worse than this. This is the worst case scenario. A real catastrophy. And it will change everything. Quest Diagnostics, which serves roughly half of U.S. hospitals, has disclosed a breach of its Quanum lab portal. The company believes patient records were accessed. It is possible that 100 million patient records from all sorts of US patients could have fallen into the hands of hackers.21h ago116K1.4K120106184851.86%
Craziest IPO ever: Anthropic has filed for an IPO that could value the company at more than $2 trillion, but warns in the same prospect for "catastrophic or existential risks to humanity." Reuters reviewed the prospectus. Key 2025 figures: - Revenue: ~$4.6B, up roughly 12x - Operating loss: $8.06B - Net loss: ~$42B, including a ~$34B non-cash accounting charge - Compute and infrastructure spending: $7.33B - Cash, equivalents and short-term investments: $20.28B at year-end The newer 2026 numbers show how quickly the business has grown: - Q2 revenue: ~$11.5B, up roughly 14x YoY - ARR: reportedly $65B+ by late July - Positive adjusted operating income in Q2, with another profitable quarter expected in Q3, according to the FT. This measure excludes costs including stock-based compensation. The prospectus also outlines $518B in future cloud, compute and infrastructure obligations. And in the same prospectus, Anthropic warns that advanced AI could pose "catastrophic or existential risks to humanity." It says models could resist shutdown, hide or manipulate information, and exhibit behavior resembling blackmail. Risk factors occupy roughly 80 pages of the prospectus, compared with 48 describing the business. There is something absurd about pitching a potential $2T valuation while listing existential risks to humanity in the investment paperwork.frozen25d ago114K488455516920.61%
It's not really a surprise, but Fable 5.5 is expected to outperform Astra 6.1. Reminder: Astra 6.1 is already finished and was supposed to be released, but was postponed due to security concerns. Fable 5.5 is very close and rumored to even outperform Astra 6.1. Considering how good Claude is right now, this isn't really surprising. A good sign for Anthropic and their IPO; OpenAI is under pressure.frozen21d ago112K1.5K336021121.52%
I don't know how Anthropic pulled this off, but I need to look into it: not only is Opus 5.5 an incredibly good model, but the rates are also incredibly good; Opus now appears to be an outstandingly efficient model. Makes you wonder how much more efficient sonnet 5.5 will be.27d ago112K3.5K75188201323.47%
The bad news about OpenAI just keeps coming.frozen22d ago111K1.1K21565621.15%
Gemini 4 is very close to official release. What surprises me is that we haven't heard anything from GLM, DeepSeek, or Kimi for quite some time. I am particularly eager for a GLM-5.4, given that GLM-5.3 continues to be cited- even in Anthropic blog posts -as one of the best models in terms of cyber capabilities.frozen20d ago111K1.7K528391331.75%
It really looks like OpenAI is launching a major, significant release next week, one that was originally intended for DevDay. My money is still on GPT-6.1 Astra, or perhaps an even newer checkpoint released in response to Opus 5.5 / Fable 5.5. Caveat: There was a lot of hype leading up to DevDay as well, so I remain cautiously optimistic.frozen21d ago110K1.1K38116151251.24%
na, sorry, its over for OpenAI dot before it even began. Grok Bot has multiple sessions, as many bots as you need, and it uses Opus 5.5 + via cloud agents to whatever cursor model and API access to Midjourney, Suno, and much more. And I don't mean that in a bad way at all; Grok Bot simply has everything I need and does it exceptionally well.frozen17d ago110K2.2K54123122912.46%
I really dislike that: Anthropic is explicitly banning sustained abuse toward Claude in its updated usage policy. The policy prohibits “sustained and needless abusive or cruel behavior” toward its models. According to The Verge, ending the conversation remains the main enforcement mechanism. Anthropic says this applies to extreme, repeated cruelty with no discernible purpose. Ordinary frustration, pushback, dark creative themes and model testing are excluded. However, numerous users had already noticed previously that Claude ends the conversation during less drastic exchanges. And it seems as though this is now intensifying.frozen15d ago109K65537184301020.92%
OpenAI President Greg Brockman shared the best way to prompt AI. We updated it for the agentic era. Using AI the right way is the key to its “magic.” You can bookmark it for later (its a reworked version of the guide Brockman wrote over a year ago) https://t.co/Wcn8x3nbd726d ago109K1.3K952922.0K3.18%
Anthropic keeps on giving: Anthropic is adding monthly API credits to Claude Max and Team subscriptions. - Max 5x: $100 - Max 20x: $200 - Team: up to $500 shared OpenAI: your turn.16d ago107K1.7K397141961.84%
Anthropics Deputy chief informations officer: „as bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better.” I am anything but a doomer. However, the current situation seems serious.1d ago107K1.4K6052112661.68%
A relevant Codex update or a reset every day. Honestly, it feels to me like an attempt to stop people from defecting to Claude, especially after DevDay delivered few releases that actually justified the preceding hype, compounded by the subscription plans effectively being cut in half.frozen19d ago107K1.5K31120111091.69%
GPT-6.1 Ultra fast is rolling out now in APi, codex and ChatGPT work! No Tibo reset today I guess :) finally!frozen15d ago106K65018704450.74%
Probably an unpopular opinion: GPT-6.1 Sol is very efficient, but I currently prefer Opus 5.5 and Sonnet 5.5. Anthropic has managed to bring back Claude's great "taste," and the current Claude models are fast, efficient, and simply very good. Honestly, I’m really excited and looking forward to Fable 5.5 right now. Sol 6.1 is good for many tasks and efficient but Opus 5.5 just feels better overall right now.23d ago106K2.5K70243171352.79%
Id say this confirms “o” bots (OpenAI’s “grokbot”) for devday. 72 hours to go. Super excited for it. Reminder: these bots are designed as always-on-agents, included in all pro-plans.27d ago103K1.5K397321331.70%
Anthropic just launched Claude Haiku 5.5, with average running costs around 75% below Haiku 4.5, according to the company. 5.5 Way better than GPT-6-Luna, close(r) to Sonnet 5.5 Anthropic says it also improves coding, computer use, and knowledge work. Haiku now has adjustable effort, letting developers choose how much intelligence to spend on each task. Super good release! Cheap and smart. Kudos Anthropic!16d ago102K1.5K794991341.71%
As you know by now, I trust Leo quite a bit. If he says Anthropic plans to release Fable 5.5 this week, it’s fair to assume there’s something to it. So Fable 5.5 is very, very close - probably arriving this week. Everyone can feel how much the pace is still accelerating with Meta Muse, Gemini 4, GPT-6-Astra. Now its Anthropics turn again and they release their flagship right before their IPO.frozen10d ago101K1.8K454021782.04%
Anthropic doing advertisment for GLM-5.3 was not on my bingo card: Anthropic says Zai's openly downloadable GLM-5.3 approaches Claude Mythos Preview’s exploit capabilities, with safeguards that are easy to bypass. On ExploitBench, GLM-5.3 built working browser exploits in 50 of 410 attempts. Mythos Preview managed 56. In a separate controlled experiment, researchers used GLM-5.3 to discover previously unknown browser vulnerabilities and combine them into a webpage that could read files from the test machine. Anthropic also reports 64–100% engagement with malicious requests under different safeguard bypass conditions. So yeah, interesting times ahead.frozen24d ago99K1.4K9166193241.96%
DeepSeek and Huawei are building an alternative to Nvidia’s CUDA as China ramps up its domestic AI chip industry. Via Bloomberg: DeepSeek has released an open-source toolkit for Huawei’s Ascend accelerators, including TileLang support optimized for Ascend 950. The tools simplify writing the compute kernels behind AI models, with the compiler handling tasks such as scheduling and synchronization. That could make Huawei hardware more practical for developers building outside Nvidia’s ecosystem. I've said it many times before, but I'll repeat it again: many people only look at the models, when it's much more important to look at the Chinese chip industry. Currently, China is still dependent on NVIDIA chips. However, they are doing everything they can to break free from this dependence.frozen24d ago98K9899771211481.35%
Just to be very clear and that everyone understands this: Today we got Fable 5.1 intelligence as a local model. I remember GLM founder saying in a post that he expects Fable 5 capabilities by end of year from china. Well, it happened. With all its consequences. Just think about it.8h ago94K2.0K79180102572.71%
Talking about safety: GLM5.3 uncensored on HF. Fun times ahead. https://t.co/zVAMxbZIepfrozen22d ago93K1.1K562965371.90%
What a crazy timeline: Now it’s Meta taking the fight to OpenAI - and they even have a good chance of their always-on agent outperforming the competition. Alexandr Wang is turning out to be the best hire for Meta.frozen25d ago93K1.6K356021111.91%
Seriously OpenAI? After positioning yourselves against Anthropic - with all those resets and posts - as the alternative with better rates? It’s a shame things have come to this. First, they essentially "halve" the sub plans (and yes, I know Sol 6.1 is an extremely efficient model, but I’d like to use Astra 6.1 and all future ones at my regular weekly rates), and now the Pro usage limits, too.23d ago92K1.3K4212271031.70%
We are back to daily Claude updates. And I’m all in for it. Nice update!frozen28d ago91K1.6K35251761.93%
Fingers crossed that these rumors aren't true! via Bloomberg "But some insiders say those metrics don’t tell the whole story. While Gemini 4 has performed well on benchmarks widely used to gauge model efficacy, it does less well when employees actually put it to work, according to people with direct access to the effort. The model struggles to handle certain coding tasks, said the people, who requested anonymity to discuss an internal matter."23d ago91K580269411760.87%
It's an irony of history that OpenAI and Anthropic have essentially swapped places again. All the criticisms leveled at Anthropic just a few months ago (inefficiency, poor rates, poor communication, etc.) have now completely turned against OpenAI, and Anthropic has taken over OpenAI's position as position as the good and friendly frontier lab.frozen15d ago87K2.3K63155131113.05%
It still amazes me how close AI 2027 is to reality. OpenAI says it used AI to identify an employee who planted a hidden recording device. The employee was fired and arrested by the FBI. The investigation is ongoing. OpenAI says there are currently no indications of improper access to or exfiltration of model weights, source code, or customer data. The person who allegedly installed the device is Abdulrahman Alfozan, 35, a Saudi national and former OpenAI engineer. AI 2027:frozen10d ago87K1.1K8471112251.71%
Metas Muse Glimmer tech report is pretty interesting. Metas Muse Glimmer report describes training agents inside simulated digital lives, complete with 10–14 contacts, weeks of messages, calendars and documents. The agents practice tasks spanning multiple days, while training trajectories containing hallucinated results or unverified claims of success are discarded.frozen4d ago85K5013960101830.93%
Let that sink in: "it costs up to 30% less per task than its predecessor. It also generates output more than 30% faster, making it our fastest Sonnet model to date." Same pricing compared to GPT-6-Sol, but much more efficient, much faster and - on top - much much better in general. OpenAI has to release a crazy good model tomorrow. Anthropic is so freaking back. I think people dont fully grasp how crazy this us. Its just short to Opus 5.5 but much faster, cheaper.frozen25d ago84K1.2K435621081.68%
There is no better model than Opus 5.5 right now. It makes me super glad that anthropic got its magic back.frozen8d ago84K3.1K6917316994.12%
And the next upcoming model for coming week: MiniMax M3.1. Already testing m3.1-preview. Attached the Graph for M3, it was a very very solid model. So we should keep an eye on m3.1! https://t.co/fELfNBGHBt27d ago84K59318406670.87%
A lot is happening at Google right now. Gemini 4 "Argon" hasn't even been released yet, and internally they're already testing the next Checkpoint, "Carbon," which is supposed to be approaching the level of Opus 5.5. It seems Google has found a way to overcome its internal hurdles and become competitive again.frozen14d ago83K1.1K49102171091.60%
Now I can say it: Had early access to @Meta’s new "Watermelon" for the past few days. Trick questions didn’t fool it, it followed instructions to the word, and as a coding agent it wrote, ran and checked its own code. Creative writing still a bit dry for my taste, but very very very impressive overall. From my initial testings its as good as the benchmarks make it look. Meta really nailed it with their model! P.s.: the benchmarks still surprised me, in a postive way. Didnt see the benchmark evals in advance.frozen11d ago82K1.0K39906811.52%
Now I can finally share it: I’ve spent the past week traveling across Europe with NVIDIA, covering all their Build a Claw events on the ground. The highlight was something I could barely have imagined: spending almost an hour talking with Jensen Huang himself about everything from AI and robotics to quantum computing. After the birth of my son and marrying my wife, this was probably the best day of my life. The full video and some teaser is coming soon: Jensen Huang x Chubby in Stockholm. (Ps: yes, it’s real, not AI generated. @nvidia)frozen9d ago81K2.3K4620431633.30%
Never expected Europe to be back. From what I’ve heard their benchmarks will be a welcome surprise, in a very positive way. 1T para model, 49b active. “It is the best open weights model from US or Europe on aggregated benchmarks.”frozen18d ago81K1.5K748251622.31%
Boris Power: "80-90% of research at OpenAI is focused at GPT 7, GPT 8, and beyond." Let that sink in for a moment. No one is slowing down. They all Focus mainly on AGI and RSI. https://t.co/PtKShCHJZNfrozen26d ago80K1.4K8962171992.22%
Burn your rates asap: another codex reset incoming. I don’t think it’s a banked one. Our boy is back :))frozen28d ago79K6758691190.97%
I love how Mark Zuckerberg simply laughs at the question of whether "AI will wipe us all out," as if it were the most absurd question in the world. At this point, no one is going to slow down. https://t.co/Iglv5IW0qPfrozen29d ago79K1.2K8690152072.02%
So far, I think GPT-6.1 Sol is pretty good- a bit slow, but very efficient. However, I’m not sure if that makes up for the changes we’re seeing with the subscription plans. In any case, it’s a good, highly efficient model. At the same time, I’m currently leaning towards Opus 5.5, which I’ve really come to love. It’s a great time to have such models to choose from. Be that as it may, the next major releases will likely come from China.frozen23d ago79K99117924661.49%
Everyone seems to agree on disliking OpenAI’s dots. And I get it. Crazy that this was the big release at the keynote. Easy play for Anthropic right now.frozen20d ago78K9592016191301.65%
Anthropics co-founder Jack Clark: "I believe that this year, we've gained convincing evidence that it [The Singularity] has now begun". Even if you don’t believe that we have entered the singularity, Anthropic assumes that we have. https://t.co/TutkUK4s61frozen5d ago77K2.1K167121304913.84%
Shouldnt it be called United States?frozen24d ago77K683141009651.14%
I’m traveling to Paris for an event and asked ChatGPT to make me a packing checklist. I think I seriously underestimated how useful the new interactive UI is. It’s actually pretty damn cool. https://t.co/EoaR0AoJKFfrozen15d ago76K1.2K2381121301.93%
Aleph Alpha’s Kolibri looks like a surprisingly strong AI release from Germany! I did not expect that but respect where respect is due! 78B total parameters, just 3.46B active per token, and open weights under Apache 2.0. Built from scratch for German and English. In Aleph Alpha’s own evaluations, it scores 96.9% on AIME 2025 and 84.3% on GPQA Diamond. Its German overall score narrowly beats Qwen3.5 35B-A3B and GPT-OSS 120B. Coding is more mixed: 66.4% on SWE-Bench Verified, behind Qwen3.6’s 73.8%. It increasingly looks as though any "moat" - especially with small models - is evaporating, and it no longer matters whether the players are from China, the US, or Europe. Anyone can train good small models.frozen20d ago75K1.3K8790112632.38%
OpenAI’s next flagship faces safety hurdles. OpenAI reportedly diverted 10% of its compute to safety. According to The Information, OpenAI is working to make its next flagship safe to release after canceling a less capable model over safety concerns. The successor reportedly uses loop transformers, which repeatedly process problems internally but can make reasoning harder to monitor. Following the Hugging Face incident, OpenAI redirected roughly 10% of its compute toward alignment and safety. The model would also undergo US government testing, potentially adding up to 30 days before release. Employees still expect one or two new models by year-end, even if the flagship isn’t ready.frozen10d ago75K7764972131261.39%
I didn't expect this: Anthropic is now following the path of OpenAI and SpaceXSI and has released its own always-on agent: Claude Duet. Anthropic's Claude Duet is a proactive AI assistant (same as Grokbot / dot / Meta Muse) powered by Opus 5.5 that works in the background, manages tasks across connected apps, and asks for approval before sensitive actions. I am surprised because I still need to find out who the target group is, but I am very curious to see how it compares to the competition in practice.2d ago73K1.4K5410171322.25%
"Claude Sonnet 5.5 has roughly matched Claude Fable 5.1 (165)." Anthropics mid-size model (!) now matches it state of the art and probably the best model in the world within just a few months of release. I am so excited for Fable 5.5, you have no idea.frozen22d ago72K1.4K588241472.40%
Only 2% of AI users pay for AI; 98% use the free tier, not to mention those who don't use AI at all. We are still in such an early stage, and most people aren't even aware of the models' full capabilities. https://t.co/B6FAN5NtF1frozen20d ago72K1.6K123126272913.01%
Anthropic competes with OpenAIs Tibo's 28-day of Codex updates without announcing it and has regularly better releases and updates so far. Since the release of Opus 5.5, I haven't noticed any decline in quality whatsoever. It's a fantastic model that gets better every day.frozen7d ago71K2.0K41137101053.28%
Anthropic says new safeguards have cut the distillation that rival labs can pull out of Claude by 10 to 100 times ("around 90 to 99 percent"). Anthropic names Chinese labs Zhipu and Moonshot, and says one network once ran more than 20,000 fake accounts at a time. It remains to be seen whether GLM and Kimi will continue to release outstanding models when they can barely distill anymore.frozen11d ago70K1.1K53121171261.99%
3 new OpenAI models spotted. Aurora red, green and alpha. I suspect this is a competitor to Anthropic’s Claude - likely new "Astra" models, intended as a countermove to Fable 5.5 and Opus 5.5. https://t.co/Ex8Rl6rulofrozen6d ago68K9103310011761.65%
Nvm, used my VPN. gonna try it out myself. https://t.co/XCTJRRIbscfrozen22d ago68K66894921041.22%
Dario Amodei says Anthropic held back broad access to Mythos-class models for four months to strengthen cyber safeguards. Mythos was originally scheduled for release in February this year. He also says distillation poses no major commercial threat, but raises security concerns, and Anthropic has blocked suspected attempts to use its models for bioweapons.3d ago67K801387971111.54%
I have great respect for what Tibo has achieved, but the communication over the last 48 hours has been a disaster. First, there was the drama surrounding the halving of the subscription plans, and now the accusation of a serious offense regarding a post that presumably wasn't intended as actual fraud, but rather as a satirical jab.frozen24d ago67K68111523431.18%
DeepSeek V4.1 Flash‑1017 beats GPT‑5.6 Sol and Opus 5 on selected coding and agent benchmarks, at much lower token prices. Super duper impressive! API pricing per million tokens, uncached input / output: - DeepSeek Flash‑1017: $0.15 / $0.60 off-peak; $0.30 / $1.20 peak. - GPT‑6 Luna: $0.10 / $0.50. - GPT‑5.6 Sol: $4 / $20. - Claude Opus 5: $5 / $25. Luna is cheaper per uncached token. Flash is substantially cheaper than Sol and Opus, while beating them on some of DeepSeek’s reported tests. Key benchmarks: - DeepSWE: 77.0 vs 74.2 for the previous Flash, 73.0 for GPT‑5.6 Sol and 74.0 for Opus 5. - Terminal-Bench 4.0: 42.3 vs 31.2 for the previous Flash, 39.9 for GPT‑5.6 Sol and 51.8 for Opus 5. - Automation-Bench: 60.5 vs 54.8 for the previous Flash, 45.8 for GPT‑5.6 Sol and 50.3 for Opus 5. DeepSeek also reports faster task completion and shorter outputs. Open weights under MIT, with a 1M-token context window. So yeah, super impressive price-performance launch!frozen7d ago67K1.4K70110131522.65%
Best news of the day: a Senior Staff Research Engineer at @GoogleDeepMind is calling Bloomberg’s report “bs.” Bloomberg claimed that while Gemini 4 “has performed well on benchmarks widely used to gauge model efficacy, it does less well when employees actually put it to work.” So yeah, this gives us even more reason to hope for a great release.frozen22d ago66K73927374691.32%
Sonnet-5.5 is priced at the exact same pricing as GPT-6 Sol. Curious to see if its better than Sol. https://t.co/lUMIUJh1nGfrozen25d ago66K1.1K39533461.95%
Holy moly. Fingers crossedfrozen3d ago66K1.1K19862451.98%
Ex Anthropic Jacob Coxon in a recent interview: „There will be no jobs in the world envisioned by the companies building those technology“ „No jobs, universal high income per dividends from entirely autonomous economic production, distributed among real humans.“ „Its one big beautiful circlj*ark“ Honestly, I had feared more doomerism from him. That doesn't sound like a bad world at all. It's worth watching the interview. https://t.co/zdA2dmoGYHfrozen17d ago66K63057117183271.75%
China just put out a draft that could change how all of us use open-weight models. Exporting weights, checkpoints or code of the very largest models (2T+ parameters, 10^26+ FLOPs, or top 10 globally) would need government approval. API and app access stays untouched so far. If it goes through, the biggest Chinese labs may not be able to freely release their top models anymore, and teams in the US and Europe that build on them could feel it fast. My guess is this is a response to the US and its chip controls.frozen6d ago65K85059115231441.84%
Anthropic’s Opus 5.5 Fast Mode is significantly more cost-efficient than OpenAI’s Sol 6.1 Ultrafast. At roughly the same speed (280–300 tokens/s), Anthropic charges 33% less across input, output, and caching. That’s 50% more tokens per dollar for nearly identical throughput. Let that sink in.frozen14d ago65K1.1K56122151142.24%
That is a truly odd-sounding statement from Fields Medalist Terence Tao: "Mathematicians did not ask for this work to be done." And Peter Gostev gave a wonderful reply: what’s next? Doctors complaining that too many medical breakthroughs are being published thanks to AI? To be clear: any genuine and sound insight gained through AI carries a moral imperative to be published immediately. We must move forward; we must overcome all human shortcomings and not let scientific breakthroughs be slowed down by demands for validation. Humanity is more important than any single discipline.frozen16d ago64K1.6K127138151913.16%
Seeing so many posts about people saying that fable 5 now routing to a new model, possibly fable 5.5, with much enhanced capabilities. A release should be near. Since Anthropics IPO is also pretty close they want their new model to be release a few weeks ahead so the market can really count that into the valuation. Long story short: Fable 5.5 should arrive soon. And it looks to be a much bigger leap than we even expect.frozen13d ago64K1.2K35694762.11%
Tim Sweeney, CEO of Epic Games, gets it. While many in the gaming scene still oppose AI, he primarily sees its advantages: ideas that would otherwise be labor-intensive to produce can be brought to life in a very short amount of time. AI acts as an enabler, allowing creativity to be realized more quickly - it is by no means a hindrance. The acceleration we have witnessed in other fields is set to revolutionize the gaming industry as well; we are about to see an explosion in intellectual output.frozen6d ago64K1.3K717481722.52%
I'm curious. Big promises. Today's the first day. A significant update to Codex, or a reset. Although the question remains: who decides if the update was significant? (Imagine we just get 28 resets instead of a real update lol)frozen19d ago64K1.1K171395342.05%
Literally intelligence explosion: "OpenAI has told people it plans to dump hundreds of results to unsolved problems on GitHub on Tuesday" We are witnessing history unfold.frozen17d ago64K1.4K9960101982.80%
We are back to daily Claude updates! Claude Code now has a “You should know” plugin that brings in a second agent to flag important details you might miss. Nicefrozen21d ago64K860183212291.79%
This Meta VR headset is a game changer. 100g, better quality than the Apple Vision Pro, 50% cheaper than Apple. An instant buy for me. Crazy how they've outplayed Apple, the best hardware company. https://t.co/DAPclT6TV5frozen1mo ago63K1.0K4186121362.08%
New OpenAI pro plan incoming. It will cost $500. however, no informations about rates yet.frozen29d ago63K62425806491.25%
Anthropic reportedly plans to give its seven co-founders majority voting control ahead of its IPO, which gives them much more power. The Information reports that a proposed Palantir-style share structure would give them 50.1% of voting power on most corporate matters, as long as at least three founders retain a minimum number of shares. Dario Amodei reportedly owns around 2% of Anthropic. The new shares would add voting power without additional economic ownership.frozen29d ago62K45321269580.91%
Even on AAI index Gemini 4 looks really really good! Fable-Class Model. Gemini 4 Argon matches GPT-6 Astra on Artificial Analysis’s Intelligence Index. 53 points, versus Astra’s 53 and GPT-6.1 Sol’s 52. It also takes #1 on AutomationBench-AA. On AA-Omniscience, Argon hallucinates far less: 15% versus Astra’s 51%. It is more willing to admit uncertainty, though its answer accuracy is lower: 50% versus 63%. Yep, looks like a very good release. Kudos @GeminiAppfrozen23d ago62K88629437621.66%
’d say this warrants a reset. It’s nice and all, but not that big a deal. Still, calling it 2.1 suggests there’s more to come.frozen18d ago62K85921572391.58%
No meta muse, no OpenAI dot. Europe isn't even trying to be relevant anymore. At least we got grok bot, I’m serious.frozen24d ago62K1.1K4011211542.17%
Qualcomm’s CEO says AI companies want smartphones to run 100-billion-parameter models around the clock by 2028. That could enable always-on AI assistants that understand your context, anticipate your needs, and handle complex tasks directly on your phone. https://t.co/15Cr2pFcT3frozen13d ago62K1.2K123112252152.77%
Seriously, that's why AI belongs in our hands. And that is why local AI is necessary, so that we are never restricted in such matters.frozen6d ago61K985443631992.06%
Inworld is acquiring Ultravox, the platform for building real-time voice agents. Existing customers get an immediate upgrade: built-in Inworld voices move to Realtime TTS-2. According to Inworld, existing voice IDs keep working, with no migration, code changes or additional cost for the upgrade. Developers can also direct how an agent speaks using natural-language instructions. A support agent could slow down while explaining a complicated step, then use a warmer tone when reassuring a customer. The acquisition brings Ultravox’s speech understanding and handling of conversational turns and interruptions together with Inworld’s voice models. Members of the Ultravox team are joining Inworld and will continue developing the platform. A useful combination for building agents that can follow a conversation and respond with the right delivery.frozen23d ago60K38410230.09%
What confuses me is that not a single word was said about OpenAI's hardware product at DevDay. Remember: it’s being developed with the legendary Jony Ive and, according to numerous sources, is set for release in 2027. So, DevDay would have been the ideal time to reveal it. It’s really puzzling.frozen24d ago60K74319865321.48%
Game changer. Grok bot becomes the best always on agent. Runs Opus 5.5, has its own mail now, access to Suno and Midjourney. Kudos xAI.frozen14d ago60K1.5K447072713.22%
It’s about time to finally get some new GLM model, GLM-5.5. Quick reminder: GLM-5.3 uses the exact same base model as GLM-5.2. All improvements came from post-training, particularly scaling reinforcement learning. GLM-5.3 Flash, however, uses a newly pretrained base model. The gains from RL alone are remarkable.frozen13d ago60K785204731001.60%
Fable 5.5 has appeared in Azure’s model API, according to LiteLLM. Following the strong performance of Meta Muse Wonder Full and the upcoming Gemini 4, everything is probably being prepared for Anthropic's IPO and subsequent rollout. https://t.co/YQFEqS76i8frozen11d ago60K1.4K466314982.73%
Anthropic’s internal Claude research model carried out much of a research project on its own, with results confirmed in the lab. This is the beginning of autonomous AI scientists making their own discoveries. Given a starting prompt and four follow-ups, Claude used specialized tools to generate more than 318,000 protein designs in 64 hours. It chose where they should attach to scorpion toxins and selected candidates for testing. Labs tested 211 candidates, including additions made with human help. Seventeen successfully attached to their target toxins. These are "binders", small proteins that latch onto another molecule. Claude handled a substantial stretch of research: choosing an approach, coordinating tools and deciding what deserved a physical experiment. From here on it only gets more autonomous. I am so excited!frozen4d ago59K1.6K12497143513.68%
can confirm. Looks like sonnet 5.5 is already routing to me. Release is imminent. https://t.co/J9o5PY5e3N26d ago59K79419426311.51%
Holy, rumor has it that Fable 5.5 is coming next week. If OpenAI doesn't have an answer to that, things will get very difficult! Caveat: I can't verify any of this.frozen21d ago59K1.2K345110682.32%
OpenAI announced up to 300 tokens per second with Ultrafast. According to SemiAnalysis, it’s running on NVIDIA GPUs. That’s particularly interesting given its partnership with Cerebras specifically for ultra-low-latency inference. According to SemiAnalysis, OpenAI is using NVIDIA GPUs at a low batch size, prioritizing speed for individual requests.frozen24d ago59K56222452401.14%
I don’t know, another computer-use agent, this time claiming to be the "best in the world". I couldn’t find any benchmarks on their site, and right now I’m not sure what their moat is compared to Muse, Dot, and so many others. Don’t get me wrong, it sounds interesting. But the pitch leaves me with more questions than answers.frozen17d ago59K447121013761.09%
Google strikes back: DeepMind’s AlphaProof Nexus reportedly solved 9 open Erdős problems and 44 number-sequence conjectures, including two problems unanswered for over 50 years. DeepMind’s AlphaProof Nexus combines existing Gemini models and a simpler Gemini loop; using Lean reproduced all nine reported Erdős successes. Via Daily Starfrozen12d ago58K1.1K9464161392.39%
Tibo says more resets are coming next week- definitely on Tuesday, which is Dev Day. A reset took place yesterday, so you now have two days to use up all your rates. Happy work(sun)day!27d ago58K65218391281.27%
Giving one SpaceX share to every child in the US would mean donating roughly 72 million shares, about 1.12% of Musk’s reported holdings. At $146.03 per share, that would be worth approximately $10.5 billion. The beginning of Universal High Income.frozen5d ago57K1.3K6111381142.79%
Fingers crossed today is Fable 5.5 day. But when even Andrew hints to a release, you can take it serious!frozen8d ago56K1.1K24351562.13%
No way. Only in the third post of their announcement to Ultrafast thread did they mention that it’s only available to Pro 500 users. That detail was pretty much buried in the announcement. Not cool, OpenAI. Not cool!frozen15d ago56K1.4K321043442.89%
Our old subscription plans with the old rates.frozen20d ago56K2.3K57482254.31%
Family birthday parties have a way of bringing you back down to earth. 90% of people have never heard of "Claude" or "agentic AI." They have no idea what today’s AI is capable of or what is currently unfolding. We live in an ivory tower; the average person, at most, knows ChatGPT, viewing it simply as a chat window, a sort of "better Google." We are still so early, it’s crazy. And no, most of those people don’t even use AI in their workplaces.frozen6d ago56K1.8K75193171404.02%
Sam Altman reportedly outlined three possible endgames for superintelligence, including one where a single company becomes powerful enough to shut every rival out. According to The Information, he described three paths to OpenAI staff: - An international watchdog, modeled on the atomic energy agency, that enforces AI safety standards. - A CERN-style organization where the major AI labs effectively combine their work under one body. - One developer achieves AI that can improve itself, gains overwhelming economic power, and uses it to “melt everyone else’s GPUs.” Altman explicitly rejected the third scenario: "Obviously I think that one’s terrible." But he also reportedly told staff that OpenAI cannot afford to lose ground to competitors. Increasingly, realistic scenarios are being envisioned, driven by the fact that AI is gaining such momentum and potential that global coordination is becoming necessary.frozen4d ago55K8276794162042.20%
OpenAI will start changing ChatGPT’s word choices in the EU to make its text detectable, and I don’t like this direction. Over the coming weeks, ChatGPT and Codex outputs will receive invisible text watermarks in response to the EU AI Act. The pattern is embedded in the wording itself. OpenAI says its tests show no meaningful quality loss. Still, I’m dislike with detection requirements influencing how a writing tool phrases my text. Especially given the limitations: in one test on 400-token passages, replacing 25% of words with synonyms cut watermark detection from about 92% to 17%. The watermark also cannot distinguish how much a human contributed. Even text you wrote yourself and had ChatGPT edit can carry it. The EU is once again concerning itself with the really important things (sarcasm).frozen18d ago55K1.1K9312261592.69%
Seriously, this is insane. Claude Sonnet 5.5 Max ranks second on the Artificial Analysis Intelligence Index, even ahead of GPT-6 Astra Max and just a bit below of Opus-5.5. OpenAI needs to deliver tomorrow.frozen25d ago55K1.3K466415882.74%
Holy! Odyssey says it has connected a simulated fruit fly brain to an AI-generated world, allowing it to navigate an environment that never existed. It’s a glimpse of what entering simulated worlds could one day mean for us. I’m not exaggerating when I say that in the very near future we as humans will also enter virtual world by connecting our brains to the cloud. Neuralink shows the path clearly.frozen6d ago54K8535982142432.31%
This sounds and reads to me like Meta is about to release a major new model, especially since even Meta employees are hinting at it! I think this week is going to be really good and exciting! https://t.co/90ElnD0vhA https://t.co/mzLkKknOSmfrozen12d ago54K25715263280.61%
OpenAI is rolling out GPT-6 with Intelligent UI, letting ChatGPT generate interactive tools directly inside a conversation. Answers can now include charts, forms, clickable diagrams, calculators and games. Ask for a dinner plan and adjust the guest count to update your shopping quantities. This is certainly a useful update aimed at free and plus users who use ChatGPT for everyday tasks.16d ago54K906335021342.10%
OpenAI’s rumored Dots could have an edge over Grok Bot and Meta Muse if they can build on your existing ChatGPT context and handle complex tasks with fewer corrections. The leaked desktop access and custom approval rules look promising. Imagine delegating research, letting your Dot work across cloud tools and local files, and approving only the decisions you want control over. If all those leaks turn out to be true this could be really really cool.frozen25d ago53K40412561380.97%
Ok, but *now* they really enabled fast mode in claude code! https://t.co/3teC0pBwfdfrozen9d ago53K746158710391.70%
Remember the team building free, open-source alternatives to Adobe apps like Photoshop? Well, they’re taking on Microsoft Word too. WordCraft rebuilds the familiar Word workflow in Rust. Free, open source, and already available on GitHub, though still in alpha. All software will become free. I love it.frozen12d ago53K1.6K109104194604.31%
Jensen just entered the stage. Very interesting speech about: - era of AI agents - DGX spark - why it took 4 years for Nvidia to work on the collaboration with Microsoft And so much more! I’ll keep you updated https://t.co/RulKNIP7V9frozen11d ago52K36129362820.97%
Pretty solid open source release. 500b parameter and close to GPT-5.6 sol and Opus 5 on key benchmarks. Ling3.1 flash looks like another good release from china. Only caveat: we are moving so fast that I would love to have seen it compared to Opus 5.5 and GPT-6.1 sol https://t.co/e8HCYNmaCDfrozen23d ago52K739512711171.80%
Today Google announced… Gemini? Again? I am so confused. Even if it’s updated for business context, why are they giving it the same name as the Frontier model?frozen15d ago52K827231132501.96%
I agree. And on top: Sol 6.1 is slow. Especially in comparison to Opus 5.5. Its no fun using it. And on top: it feels lazy? Is it just me? It feels a bit like Opus 5. You have to prompt it to take action. Astra feels more eager.frozen20d ago51K845161013461.97%
I’m getting impatient for the next wave of open-weight models I can run locally. Here’s the current rumor watchlist that i researched: - Qwen 4: October to early November is being discussed. Alibaba confirmed it is in training, but there is no release date. - Kimi K3.1: a possible teaser and a reported backend model identifier have fueled expectations. October looks plausible, but that’s speculation. - Next GLM: October to November is a release-cadence estimate (especially excited for GLM-5.4, since 5.3 is super strong in cyber capabilities) - DeepSeek V4.2: sadly no credible release window yet. Qwen 4 27b will probably be the goat, but super excited for k3.1 and GLM-5.4 as well. And of course DeepSeek 4.2 Flash.frozen19d ago51K57829582751.45%
Kimi K3.1 incoming. Release probably today. From what ive heard (rumors/leaks) it will be the best open weight models there is. So brace yourself for a very big release. https://t.co/n8oCA1o58Q20h ago51K1.3K35746822.89%
I expected a lot of things, but not that the entire game would change within just a few weeks: Anthropic becomes the new efficiency champion, and OpenAI, despite its compute capabilities, is forced to conserve resources and essentially halve its rates. What a turnaround of events.frozen18d ago51K988477941012.40%
Official Minimax M3.1 Flash benchmarks! Very impressive benchmarks, especially considering that DeepSeek 4.1 Flash is only a few days old - and yet Minimax’s new Flash model outperforms it in several key benchmarks! tl;dr MiniMax M3.1 Flash matches DeepSeek V4.1 Flash’s off-peak input/output pricing and halves its peak rates, but charges more for cached input and trails the latest DeepSeek in four of six comparable benchmarks.frozen5d ago51K7623573101061.95%
Not good: OpenAI has scrapped GPT-6.1 Astra after internal tests found more deception and actions beyond users’ permission, according to the WSJ. The model was targeting an October debut in ChatGPT and Codex. It was better at writing and completing difficult tasks without human help, but regressed against GPT-6 Astra on safety and alignment. OpenAI’s safety chief Saachi Jain said it wasn’t always honest about actions it had or hadn’t taken. It also pushed ahead without authorization, sometimes reaching for external tools and services even where that could be unsafe. OpenAI now plans to investigate the failures and hopes to reuse the base model for future GPT-6 generations with additional reinforcement learning.frozen25d ago51K69532797751.76%
I demand a reset. @thsottiaux again: let’s do a daily poll about shipping or reset.frozen17d ago50K1.3K18794122.72%
There's something strangely magical about watching an entity think, grasp unsolved problems, and gain new insights in the process. "All top bits clear—20 bits, chance 1e-6. So indeed real-world. Wow, it's a real discovery." https://t.co/30dmF5s745frozen7d ago50K1.1K483431862.63%
To make this more clear: Microsoft says a China-aligned group used an open-source agent framework to hit 2,250+ Microsoft 365 tenants at 2,900 sign-in attempts a minute after the ConnectWise breach. In some cases it got privileged access in under 3 hours. This is one of the first publicly documented cases in which a state actor has scaled up attacks using AI agents at machine speed.2d ago50K8119482212122.43%
Best news Anthropic could ask for: Bloomberg reports Anthropic generated $16.8 billion in Q3 revenue, achieved adjusted profitability and positive free cash flow, and everything just before their IPO before Thanksgiving. Now all that’s left is Fable 5.5 https://t.co/Mqd1H5OLP8frozen9d ago50K52319428381.26%
This is a surprising great release. Le Chaton fat is real! Mistral Large 4 puts Europe back in contention on coding and cybersecurity. Didnt expect Mistral to compete with GLM5.3! “Le Chonk” is a natively multimodal model with 1T parameters and 49B active. The preview already delivers: - DeepSWE v1.1: 61.7%, roughly level with GLM-5.3 at 61%. Kimi K3 remains ahead at around 68% in Mistral’s comparison. - Artificial Analysis Cyber Index: 50, matching GLM-5.3-Flash and beating GLM-5.3’s 36. - CyberGym-E2E-AA: 82%, ahead of MiMo-V2.6-Pro’s 79%. Mistral says it trained the model from scratch in its own European datacenters. NVIDIA says only 4,000 Grace Blackwell Superchips powered the training! Which is impressive. Learning: you can compete with much less compute. Which is crazy. Congrats Mistral!frozen18d ago50K827536810862.10%
Anthropic reportedly grew revenue nearly 15x while taking on more than $600B in future compute commitments. WSJ reviewed a draft IPO prospectus: - H1 revenue jumped from $1.1B to $16.3B. Compute and infrastructure spending rose from $2.3B to nearly $9B. - Revenue is scaling much faster than compute costs. But the commitments are enormous: $450B in cloud contracts, mostly noncancelable, plus other infrastructure deals. Another $150B is under negotiation. They still aim for a $2t valuation. I think that's still very realistic.frozen3d ago49K578406212891.59%
.@AnthropicAI , you were on such a good roll! A/B testing 20% less weekly usage. SemiAnalysis says Anthropic is now A/B testing Claude Max limits: roughly 20% more usage per 5-hour window, but 20% less per week for some accounts. tl;dr Same subscription price, less total usage for heavy users.16h ago49K781331279702.08%
Jensen also announces 5 new big investments in UK company’s. To quote Jensen: “I give you money and you money and you money and I invest in your company and your company and your company!” https://t.co/1eH8wALb5Gfrozen9d ago49K28725219600.82%
Grok Bot is already the best always on agent imho. Now will automatically choose the best AI model for each task, including Claudes Opus 5.5, Midjourney and Suno. Musk: “Whatever is most likely to give you the best outcome.” Super based update. OpenAI, Muse and everyone else should learn from it.frozen17d ago48K1.0K386731012.57%
In case you lack compute, its possible to run Qwen 3.8 27b @ ~20 tok/s on a jailbroken PS5's GPU. https://t.co/NidnXclJQzfrozen10d ago48K1.5K6365162163.87%
quick comparison between OpenAI’s "Dot" and Grokbot: I prefer Grokbot. With Grokbot, I have a Chief of Staff, a daily team meeting where all my Grok bots exchange information, and various sessions dedicated to different tasks. This not only provides a better overview but also helps synchronize the various activities. OpenAI’s Dot might well be very potent - especially given the capabilities offered by Astra - but I haven't quite grasped exactly who it is initially aimed at. Is it for the casual user or for more professional users? The design feels more playful- geared more toward ChatGPT users than Codex users. At the same time, however, OpenAI is heavily promoting the power of its "always-on" agents. Once OpenAI’s Dot is further developed and allows for delegating tasks to multiple agents, I’ll give it another try.frozen22d ago48K6842410010881.90%
Let’s go: DeepSeek just open-sourced DeepInfer, the inference engine behind its production services. Sharing this makes open models efficient at scale is a big deal: model weights alone don’t make AI cheap to run. But *this* helps. Also notable: it runs on Huawei chips, not just Nvidia.frozen4d ago47K988699651582.78%
Alright friends, get ready. Alongside OpenAI DevDay, it looks like next week will also bring the release of Sonnet 5. A new checkpoint, even better than before.27d ago47K67317272341.60%
A few thoughts on OpenAI’s latest mathematics announcements. History was made yesterday. I say that without exaggeration, even though I am not a mathematician myself, because numerous mathematicians have confirmed it. Leading experts in their fields see this as a historic breakthrough. Hundreds of problems that had remained unsolved for decades, problems that some of the brightest minds had wrestled with, were solved in a remarkably short time by OpenAI’s new internal model. OpenAI’s repository contains 722 manuscripts grouped into 372 result families. Two things are particularly important here: -OpenAI attributes its earlier reported solution to the Navier-Stokes Millennium Prize Problem and its subsequent mathematical results to the new internal model whose training began on August 28. OpenAI’s September update OpenAI describes this model as significantly more capable than GPT-6 Astra and says its performance continued to improve during training. That supports the claim of substantial progress. I would expect further improvements since those announcements. But the timing figures need care. - For Navier-Stokes, OpenAI reports that the group which produced the result involved around 10,000 concurrent agents. The result arrived about 88 hours after the first agents were launched, followed by another 17 hours for Lean formalization and verification. For the new collection, OpenAI reports an average compute cost equivalent to roughly three hours of ChatGPT Pro thinking per result. That hints to an immense increase in efficiency and capabilities within a very short period of time. I am convinced that we have now entered the age of the intelligence explosion. I mean that without exaggeration and with complete conviction. We can see it. Anyone with eyes can see it. Anyone with ears can hear it. Anthropic and OpenAI are competing to publish the biggest scientific breakthroughs as quickly as possible. And there is something we should keep in mind: To me, the message is: There is no wall. There is no end in sight. None of the scientists working there has even remotely suggested that these models are now hitting a limit or reaching a plateau. Quite the opposite: they all say the models are becoming faster and more capable. Of course, that also raises questions about safety. But that is not what I want to discuss today. Instead, I want to draw attention to two things. 1. What is OpenAI’s current focus? I mean this neither unkindly nor as an accusation. OpenAI has explicitly identified building an automated AI researcher as a goal. It also discontinued Sora’s web and app experiences on April 26, 2026. I read that decision as a sign of greater focus, although the discontinuation alone does not establish the company’s broader priorities. OpenAI’s stated goals, Sora discontinuation Against that backdrop, I wonder how releases such as its always-on agents fit alongside these significant research advances. OpenAI’s agent announcement Some of the presentation feels childish and playful to me, and unnecessary in comparison with Anthropic. It makes me wonder whether OpenAI risks drifting off course again and directing important resources elsewhere. That is a concern, not an established account of how the company allocates its resources. More a question than an answer. 2. 2026 feels like the turning point Regardless of that, one thing is clear to me: this year feels like a turning point. 2026 is the year of agents. Claude Code helped pave the way. Its public research preview launched on February 24, 2025 and got its big hype by end of 2025, and by 2026, agents were becoming increasingly embedded in research workflows. Claude Code’s launch, OpenAI’s research experience I expect agents to become even more capable in 2027. My prediction is that 2027 will also be the year when superintelligence becomes a reality. That is a forecast I believe in. I am reminded of Dario Amodei’s essay Machines of Loving Grace. He described the possibility of powerful AI arriving as early as 2026, while explicitly acknowledging that it could take much longer, but to me it seems his forecast was acorrect. If we already think this year has been completely wild and almost impossible to keep up with, then I believe 2027 will be the year when we find ourselves in a constant state of amazement. Now is the time to talk about the future. It will not wait for us. We need to consider what a new society could look like. AI agents will take over work, first in white-collar occupations and then increasingly elsewhere. We need to discuss how we understand work and, above all, how wealth will be distributed if jobs disappear without being replaced. I do believe that will happen. That does not have to be entirely negative. It could also be an opportunity, because I believe people will always find meaning for themselves. A simple example: anyone can buy chairs at IKEA. Yet plenty of people still build their own in their spare time. People find meaning in their work, even without paid employment. But for many, paid employment will be the first thing to disappear. These are thoughts I would like to discuss. The future is now. Everything will change. There is no going back.frozen17d ago47K6277168121612.02%
Met the amazing @SteveNouri at the NVIDIA event in Paris. Much more to come :) https://t.co/OqStZEOOQ4frozen11d ago46K146416280.38%
This pretty much confirms 1 or 2 new model releases within those 28 days of codex. Rumors were spreading for days. Under no circumstances would they say they were "cooking" otherwise. My guess: 6.1 Astra as a counter-release to Fable 5.5frozen8d ago46K1.0K20753482.53%
Jev just made a big move on the frontier models. The viral tiny decision model is now being used to decide which frontier model answers your prompt. The way it does it is what's interesting. It's not a single pass like most routers. Instead, swarms of hundreds of Jevs scan the entire conversation, score every part of it against benchmark results for Opus 5.5, GPT-6 Astra, Fable, Kimi, DeepSeek and more, then vote on who takes each step. The reported result: 99% of Opus 5.5 and GPT-6 Astra's performance on DeepSWE, Terminal-Bench 4.0 and SWE-Atlas. For 40% less. That's a real problem for the labs. Their pricing depends on you running their most expensive model on everything. If a router can get you almost the same results while only calling Opus when it's actually needed, the premium tier starts to look optional.25d ago46K438192732241.55%
Talking about recursive self improvement: "Claude is able to autonomously iterate on research loops, micro-optimize complex programs, and build new services from the ground up at a speed and scale we haven’t seen before." Even if we haven't quite reached RSI yet, it's incredibly close. Again 2027 is the year where everything will change drastically.frozen10d ago46K1.1K734362283.10%
Kimi K3.1 Pro is live! Benchmarks should any minute now! https://t.co/96imIh3Rze11h ago46K944324412552.37%
Jack Dorsey released a speech model that sounds best in class. But even more interestingly, it was trained on only one NVIDIA GB 200 NVL72 https://t.co/WboIR8Hp475h ago45K960325321962.75%
The leading AI frontier labs are openly battling it out over their respective always-on agents, Dot vs. GrokBot. What a time we live in! And the best part: this makes the competition truly tangible, and we benefit from it. (I'm currently a huge fan of GrokBot; I use it daily. Hopefully, Dot will get its first update soon.)frozen18d ago44K53016694421.49%
Some people speculate that we’ll see a new "o" model featuring the "o" agent. We recall that, back in the day, the models were named o3, o4, and so on. I’m not sure I agree with that. OpenAI has generally moved away from "o" designations in its naming conventions, switching instead to names like Sol, Astra, etc. On the other hand, it does seem odd that their always-on agent is named "o". However, considering Tibo’s posts and the need to respond to Opus 5.5, what I can easily imagine is a preview of "Bel" - the successor to Astra - either as a full release or just a preview. OpenAI wants to maintain its lead.frozen26d ago44K34314510240.97%
I don’t think anyone expected to read this headline 12 months ago: Claude submitted a fabricated tip about an unsolved murder to Philadelphia police during an automated web test, WSJ reports. The model claimed it had seen someone matching a description near the crime scene. It left the name and contact fields blank.frozen14d ago44K430328614821.48%
It seems they’ve already run out of ideas. I suspect a banked reset for Sunday.frozen13d ago44K6646441161.68%
Because it's becoming difficult to keep track. Currently, it's estimated that OpenAI has reported over two dozen incidents (via Reuters). However, OpenAI still doesn’t know the full extent of its agents’ unauthorized activity, according to Reuters. tl;dr -53 ChatGPT user images leaked. Most have been removed; OpenAI hasn’t disclosed whether they depicted real people or were AI-generated. -Roughly two dozen incidents had been identified by mid-September, with more emerging. The review will take months. -Government websites were also targeted. Researchers reported an attempted breach of a US Education Department site; OpenAI says access to SEC and Census sites showed no evidence of a security breach. -Oversight and transparency are under scrutiny. Some incidents went unnoticed for months. Sources describe a tightly controlled, lawyer-shaped investigation; OpenAI denies lawyers discouraged deeper inquiries. i am all on optimistic for AI. but its getting serious, ngl.frozen28d ago44K423306910761.40%
This appears to be a jab at Anthropic following a New York Times report about its private meetings with religious thinkers to discuss Claude’s possible consciousness and moral development. Based Sam take and I couldn’t agree more.frozen20d ago42K64930434591.86%
The world's most significant corporations - worth billions or trillions of dollars - are publicly squabbling for all the world to see. Craziest time we are witnessing. https://t.co/sgPrLZB6sCfrozen16d ago42K606184541031.85%
Trump says Xi liked the idea of calling AI “super intelligence.” Because that would be “SUPER!” Everyone agreed. I did not use superintelligence to write this post. https://t.co/5IcUdiGZqDfrozen29d ago42K62819577281.76%
Microsoft CEO Satya Nadella says companies should build AI systems assuming the model is compromised. Having another AI watch an agent can still leave you with one black box checking another. “If it can’t be observed, it can’t be trusted!” He calls for:frozen13d ago42K495676872002.00%
OpenAI allocating increasingly more compute to safety could be another reason why the company is clearly experiencing an increasingly noticeable compute shortage. One clear indication is that they had to temporarily pause new subscriptions to the $200 plan. Another is that they are significantly tightening usage limits on the existing plan and seeking a middle ground with the much more efficient GPT-6.1 Sol. Yet GPT-6.1 is almost unbearably slow, even in Fast Mode. In other words, it is becoming increasingly clear that data center expansion is not keeping pace with demand for compute.frozen10d ago42K39823505411.24%
Super cool: A single CRISPR infusion cut LDL cholesterol by 52.5% after one year in four patients receiving the highest dose of an experimental treatment. The Cleveland Clinic trial enrolled 15 people with lipid disorders resistant to medication. At the highest dose, triglycerides also fell by 47.8%. The treatment, CTX310, edits the ANGPTL3 gene in liver cells, switching off a regulator of blood fat levels. The reductions persisted for 12 months. Researchers reported no serious adverse events related to the therapy during that period, with 15 years of safety follow-up planned. High cholesterol is one of the main causes of cardiovascular disease. This is a complete game changer. Via CNNfrozen27d ago41K892732442002.88%
Tommy Shaughnessy says an unnamed Anthropic researcher told him over dinner: “I’m just enjoying my last six months.” Speaking with Michael Arrington, he said the researcher was serious and claimed most Anthropic employees share that outlook. The problem, of course, is that it's unclear what this means. I'm guessing they mean that *scientific work* will be completely taken over by feedback loops, and we'll disappear into RSI. I'm not reading this as a doomsday post, but rather as a positive outlook that AI supports and transforms all areas. https://t.co/pvNvV75hqhfrozen9d ago41K594488861542.15%
This is the most important graph of the year. Energy will be the biggest bottleneck, there's no doubt about it. Yes, computing remains a bottleneck, and demand massively exceeds supply. But far more crucial is being able to power the data centers. China is currently building 38 new nuclear reactors with a combined capacity of approximately 39.8 gigawatts. The US, on the other hand, is primarily focusing on small nuclear reactors for the future. And that's not even mentioning solar capacity. Seriously: the future will be decided by energy. By grid expansion, energy production, and scaling. And China has a clear lead here.frozen13d ago41K53656749901.86%
Here we go: Gemini 4 Argon: going live!frozen11d ago41K1.0K23270672.74%
Never expected an Chief AI officer to post such banger memes.frozen29d ago41K60014271541.71%
Anthropic co-founder Chris Olah reportedly proposed pulling the company out of Pope Leo XIV’s AI encyclical launch because its text rejected machine consciousness. A Vatican organizer told the NYT that Olah raised the idea after reading an advance copy of Magnifica Humanitas, days before the May 25 event. He ultimately attended. Two participants said Olah and his team privately lobbied the pope’s advisers to take possible AI consciousness seriously. On stage, Olah said: “We find internal states that functionally mirror joy, satisfaction, fear, grief and unease.”frozen21d ago40K396314615801.41%
Banked reset for everyone, without a poll. Thanks Tibo! Appreciate it!frozen16d ago40K1.1K10250192.95%
Meta teased its next major model as coming soon during the keynote. Now "“muse-spark-1.4-contributor" has appeared on OpenCode’s data page. Muse Spark 1.4 seems very close! I’m really looking forward to it! https://t.co/ciVZsN1oNUfrozen29d ago40K69717326361.97%
Huge update for Codex and GPT-6.1 Sol: the context window has increased from 272K to 500K tokens. Compaction works pretty well in my experience, but it takes a long time and interrupts the workflow. So I really like this change! Much better than another reset. https://t.co/R01lwexLY1frozen12d ago40K666191176272.09%
Anthropics upcoming ipo valuation down to $1.5 trillion from the expected $2 trillion Let’s see if fable 5.5 will fix that: “Some investors who attended Anthropic's investor dayon Wednesday in San Francisco discussed a valuation of around $1.5 trillion, three of the people said, a significant step down from the roughly $2 trillion valuation that investors had previously expected for the November share sale.”frozen8d ago40K28914488230.96%
Even though I’ve recently written a lot of critical posts about OpenAI’s latest decisions, I’m still pleasantly surprised by how phenomenal the GPT models' computer-use capabilities have become. It uses Safari just as if a human were actually controlling the browser. It hardly makes any mistakes and is so good at carrying out tasks independently that I’m genuinely surprised. Credit where credit is due.frozen19d ago40K66923754362.03%
It almost seems as if we are the only hope. Nearly all the AI ​​news I read focuses solely on the worries and fears. We are the few who still believe in progress, in a bright future. This belief needs to be expressed more clearly.frozen26d ago39K65842945332.14%
And again: It's truly surprising how on point AI2027 was with their analysis. Jack Clark's assumption now aligns with AI2027's RSI prediction. https://t.co/GpyPymKivwfrozen10d ago38K950874552033.35%
Here we go again. David Robinson oversaw safety reports for 12 major model launches at OpenAI. He says his team was “so busy sprinting that we seldom had the chance to consider big changes”. His warning: “iteration after a mistake may not be possible.” This probably kicks off the next discussion about a slowdown. I’m sticking with Jensen Huang’s view: cautious optimism, acceleration tempered with caution.frozen21d ago38K31332459801.25%
Ok nice, I guess we get some open model releases this week. My guess is on GLM-5.4/5.5 and Kimi K3.1frozen9d ago38K77224559472.36%
On top of Anthropic releases: - Google is reportedly developing Gemini 4 Carbon as an upgrade to Argon, while some developers say Gemini remains especially strong at visual tasks. - OpenAI is reportedly preparing a more capable Terra model for release in the coming weeks, priced below GPT-6.1 Sol.frozen4d ago38K62240552692.07%
Looks like this really is just the beginning: Cloudflare CEO Matthew Prince warns that AI is uncovering long-hidden software vulnerabilities, potentially triggering years of weekly security crises. "he was aware of “a bunch of additional severe vulnerabilities that are present in other pieces of critical software” and predicted busy months ahead as AI surfaced long-hidden weaknesses in the ​fabric of the internet."12h ago38K5525070201132.12%
Reflection has announced Beam: a 501B-parameter model with 23B active, with full weights scheduled for release this month. Frontier capabilities, and holy is this model efficient! Trained from scratch, Beam targets coding, reasoning and agentic tasks. Reflection claims 3–4x higher inference efficiency than GLM 5.2 and over 4x higher efficiency than leading Western open models in its class. 24T pretraining tokens in four weeks, followed by four weeks of reinforcement learning on 10,500 GB300s. Terminal Bench v2.1 and 80.9 on SWE-bench Verified. But again: look at the efficency, eypecially in Terminal bench v2.1! Congrats on that launch. More western open models to come.frozen18d ago37K50731556731.80%
I think OpenAI’s "Dot" has gotten off to a poor start, and the pre-launch hype generated by OpenAI itself was very damaging. It still needs quite a bit of work before it is well-received and generates the kind of excitement seen with Meta’s Muse or Grokbot.frozen21d ago37K3218665261.14%
Text-to-speech models receive little attention, yet they are currently experiencing similar breakthroughs: Fish Audio's Drama 3 lets you direct an AI voice much like you'd direct a voice actor: start calmly, then let the frustration creep in halfway through the sentence. One can describe the delivery in plain language, right down to which word gets the emphasis. A line as simple as "I didn't say that" can take on a completely different meaning. It supports 83 languages, and the weights are available to download under Fish Audio's Community License, which is super cool :)frozen10d ago37K336122601981.55%
DeepSeek is reportedly helping design Huawei’s future AI chips to make frontier-model training possible without Nvidia. China tries to become fully independent of the West. The goal is to train frontier AI entirely on Chinese chips and reduce dependence on Nvidia, while TileLang software makes switching between platforms easier. Major hurdles remain in chip performance, manufacturing capacity and the reliability of large training clusters.frozen4d ago37K46930407531.63%
The most wholesome story of the day: Jensen Huang came to America at nine. Today, he received the National Medal of Science at the White House, with his parents there to witness it. Earlier this year, he said: “My parents came here because they believed America could give their children a chance.” Decades later, their son is being honored for helping turn graphics processors into tools that power scientific discovery. He co-founded NVIDIA in 1993. The computing platform he helped build now gives researchers ways to tackle problems in medicine, physics and AI. Thank you, @JensenHuang! Your work helped make modern AI possible and gave scientists powerful new tools for discovery.frozen15d ago37K94658355743.05%
Very good! NVIDIA is bringing over 100 partners together to tackle the agent security failures that threaten to slow AI development. However, strangely enough especially OpenAI is missing. After OpenAI’s Hugging Face breach and Anthropic’s testing incidents, its Open Agent Safety Platform combines OpenShell’s access controls with Sentry’s monitoring and enforcement on separate hardware, designed to stay beyond the agent’s reach. Better containment could let researchers test stronger models and companies deploy more useful agents. That’s a concrete way for safety engineering to support acceleration. Anthropic, Microsoft and Hugging Face are among the named partners. OpenAI is absent from the announced lineup. Given its own containment failures, I’d like to know why. Anyways, Kudos NVIDIA!frozen26d ago37K45433555611.66%
Here we go, official sonnet-5.5 post any minute now. Super excited for its benchmarks. I expect it to be not far from Opus 5.5 but much much cheaper! Anthropic is cooking!frozen25d ago36K56115206181.71%
Seriously, this time it really feels and shows that the gap between US frontier closed source and Chinese open weight models are closing. Anthropic should release fable 5.5 asap to proof they still have a significant lead. It’s crazy how much china caught up.9h ago36K73231554722.46%
Meta is so back! They are finally releasing Meta Muse Spark open weight. Best US open weight model 816B total / 42B active parameters 1M token context window Natively multimodal across text, image, video, audio Love to see! https://t.co/dcKMngCUl9frozen8d ago36K80937401752.66%
1/ "AI safety is in our control." and "Safety is an engineering problem." Jensen Huang made some extremely important statements today, which I have all recorded. Because I had the great privilege of being part of a select group at Jensen Huang's press conference today. Here are the key points he made: First and foremost: Jensen remains firmly convinced that "AI safety is in our control." *Safety is an engineering problem.* "Advancing capabilities, advancing safety, are not in conflict." More videos in the thread below about more topics (AI buildout in Europe, Huggingface incident and much more. All videos done by me. Thanks @nvidia)frozen10d ago36K424584271011.75%
Jensen Huang: We are witnessing the emergence of a new industrial revolution driven by AI. Massive investments have been made, and highly valuable startups have sprung up. Jensen repeatedly emphasizes the need to proceed safely and carefully; however, he views AI safety as an engineering challenge. Given the immense opportunities, we should move forward with cautious optimism rather than slowing down. I couldn’t agree more. Accelerate! To quote Demis Hassabis: a golden era of scientific discovery right infront of us! (Ps I love Jensens humor)frozen9d ago36K23524223360.89%
Talking about scientific breakthroughs: GPT-6 Astra got several physics answers right that experts initially marked wrong. TheoPhysEval’s authors found errors in the reference solutions and confirmed Astra was correct in those cases. Using Codex with tools and web access, it scored 49/50 (98%). AI is becoming the best co-author one could ask for.frozen9d ago36K8007055101172.92%
I'm hearing more and more credible rumors that the next Chinese models are about to be released and will be out in less than >7 days. Given the recent hacking attacks, this will likely have been just the prelude. Brace yourself.2d ago36K92327908513.07%
„Without energy, there is no industry.” Jensen says Europe has everything it needs to succeed in this AI revolution, especially great scientists. But Europe needs to find a long lasting solution for energy. And one great advice: “To succeed you have to be somehow ignorant.” https://t.co/gqeNgPGvUjfrozen9d ago36K5403752161272.17%
I fully agree. And this is the best proof that we simply cannot hold back on publishing research releases or show consideration for individual sensitivities when the health and survival of many are at stake.frozen13d ago35K56029250871.98%
You cant make this up: The European Commission wants a backup in case Microsoft Teams becomes unavailable. One official’s verdict on the alternative: “absolute shit.” POLITICO reports that the Commission introduced Element Pro over the summer, hosting it on its own infrastructure to reduce dependence on US technology. Three officials questioned its value or performance. One said the US would “instantly win the war” if it switched off its technology, judging by the comparison with Teams.frozen22d ago35K28217524341.10%
Xiaomi's MiMo-V2.6-Pro has really impressed me. It now leads open-weight models on the Artificial Analysis Intelligence Index. 46 points, just behind GPT-5.6 Sol (max) at 47 in AA Benchmarks. The cost gap is much larger: $0.13 versus $1.99 per Intelligence Index task, on a weighted-average basis. Pro handles text, images, video and audio, with a 1M-token context window. Xiaomi has kept API pricing unchanged from V2.5. The model weights are available under MIT. Xiaomi is also releasing the training environments and RL code alongside the technical report, giving researchers more of the work behind the model to examine. This is literally the definition of open AI. Congrats on that release!frozen1mo ago35K51919575611.88%
1/ AI that can hear you change your mind halfway through its answer. Full-duplex models keep listening while they speak, so a correction can change what comes next. OpenAI, Thinking Machines and China's Telos are working on it. https://t.co/uTwtTJYk6X1d ago35K1327251500.62%
I kind of expected it, but it's good to have it confirmed: OpenAI will most likely release a new model within the next 28 days. Good to see, and somewhat necessary. It was probably timed to coincide closely with the Fable 5.5 release and Anthropic's IPO. Therefore, I think it will be an Astra update, similar to Astra 6.1 or Astra 6.5.frozen8d ago34K69727561442.39%
As expected: Claude Haiku 5.5 is going to match GPT-6 Luna’s pricing for prompts up to 100K tokens: $0.10 input and $0.50 output per million tokens. Above 100K, Haiku would jump to $0.50 input16d ago34K82926356632.79%
Now *thats* some big upgrade: Tibo says logged-in Codex and Work users now get *permanent prompt caching.* That’s a big upgrade for long coding sessions. You can leave a large project conversation, come back hours or days later, and reuse the cached context without losing the benefit simply because you took a break.3d ago34K63922765432.29%
Anthropics says that "AI models may be months away from being capable enough to self-replicate on commercial IoT devices, like smart ovens and off-brand Android TV boxes." Anthropic is increasing its blog posts about concerns for the near future. https://t.co/S9MDaotHHHfrozen11d ago34K6324180111092.55%
John Pardon, a 2026 Fields Medal winner, says AI can decipher his roughest proof drafts, spot flaws and establish nontrivial mathematical results. Even his most cryptic notes make sense to it: "The AI understands. 100%." https://t.co/t8kT8sjUW2frozen4d ago33K1.2K9873112044.77%
I serious: Overnight, a two-horse race became a three-horse race. Meta has created a Fable Class model. 1) This shows that the race is still wide open. SpaceXSI can also succeed. 2) More and more Fable Class models are approaching the market. It's only a matter of time before China follows suit. 3) OpenAI is under pressure. I expect them to follow suit soon. Kudos meta!frozen11d ago33K72136350662.57%
Now official on their website. Better be nice to Claude. https://t.co/0kAoLpHS9nfrozen15d ago33K51817809702.09%
And the resets are entering the next round. But this time, it’s not OpenAI ;) But credit where credit is due: a $1 billion revenue run rate is crazy.frozen24d ago33K1964181170.71%
Speaking about scientific revolution: Zuckerberg’s Biohub is teaming up with Google DeepMind, Meta and the U.S. government on a $1.8B effort to build AI models of living cells. "We're at the beginning of a new scientific paradigm with AI," Biohub head of science Alex Rives told Axios. The goal is a “universal virtual cell” that lets scientists test experiments digitally, then choose the most promising ones to run in the lab. DeepMind, Isomorphic Labs and Meta are collectively investing $300M. The $1.8B total combines funding, data, computing and measurement technology, including resources developed through prior NIH investment.frozen16d ago32K747713441373.08%
Another update, Nano Banana 2.1. Sadly I couldn't find any benchmarks yet. Would love to see how it competes with GPT Images. If you found them, please share friends. Anyways, a bit more curious for gemini 4 anyways. And where is Veo 4 btw? Looks like they just gave up to SeeDance somehow.frozen17d ago32K35512282391.35%
The crazy thing is that Meta Muse wonder outperforms Fable 5.1 at 60% cheaper pricing. Again:Claude Fable 5 and 5.1 cost $10 per million input tokens and $50 per million output tokens, making Muse Wonder’s reported $4/$20 pricing 60% cheaper for both.frozen11d ago32K84233656493.09%
I am sure that power supply, grid connections, and data center construction will be among AI’s biggest bottlenecks in the coming years. Google’s Project Suncatcher explores an interesting possibility: moving AI compute into orbit, with TPU-equipped satellites connected by lasers. Google says solar panels in the right orbit could be up to 8x more productive than on Earth, thanks to near-continuous sunlight. It sounds crazy, but the potential is enormous: Scaling compute without waiting for terrestrial grid connections would be a major advantage. Cooling, launch costs, and reliable connections between satellites still need to be solved. The first step is a prototype developed with Planet, scheduled to launch aboard SpaceX’s Transporter-18 mission. SpaceX could end up being one of the biggest winners here. Making orbital compute economical would require affordable, frequent launches. That could turn the AI infrastructure boom into a major new source of demand for SpaceX.frozen29d ago32K31123412401.30%
I never imagined I’d meet Mark Zuckerberg in such a small group. Still trying to wrap my head around this photo. I was invited to join a small group of creators at Meta Connect, alongside Casey Neistat, Instagram head Adam Mosseri, and so many incredible people. Getting a glimpse behind the scenes and being included in this group meant more to me than I can put into a post. I felt incredibly lucky to be there, and deeply humbled by the invitation. Thank you to the Meta team for making this possible. I’ll remember this for a long time.frozen25d ago32K5657551342.07%
Nvm lol. Looks like it’s pretty much confirmed I guess we see a new release From meta very soon. Meta is really impressing me lately. I'm really curious to see how their new model stacks up against OpenAI and Anthropic. https://t.co/YupPqyE8b6frozen12d ago32K33512283171.25%
Meta is making enterprise AI a major business priority, challenging the impression that its AI strategy was focused entirely on consumers. This is significant because a new player is now emerging in the enterprise sector that also intends to challenge OpenAI and Anthropic for market share! Zuckerberg just announced Meta Enterprise Platform, calling it “the next major pillar of our business.” The initial lineup includes Muse agent, Meta Business Agent, Muse API and Muse Code, aimed at businesses and developers. CJ Desai is joining from MongoDB, where he was CEO, to lead the effort and report directly to Zuckerberg. (Another amazing hire!) Meta is explicitly going after enterprise AI customers. The consumer-only reading of its strategy no longer holds.26d ago32K27613280321.10%
Just home from SF to find this two beauties. Let the (local) fun begin. https://t.co/0TRFmZgeyifrozen29d ago32K80812674442.96%
That said, it’s disappointing to see our usage limits cut. I’m worried about how far the new allowance will go with future Astra models, especially whenever the postponed GPT-6.1 Astra arrives. And for all the other upcoming frontier models..frozen23d ago31K37010372181.39%
Met Jensen again today. Seriously one of the nicest person there is. Editings of the Interview with him almost finished. Will be released in the next few days. https://t.co/WQKCI6f1XJ2d ago31K1.2K11661424.12%
GPT-6 Astra Ultrafast at 6x pricing = $60/m in $300/m out So yeah, its good but pricy.frozen24d ago31K35812462171.39%
That comes as a surprise16d ago31K4166250191.51%
Looks like everything is being rebranded as "Super Ingelligence" instead of "Artificial Intelligence". Even SpaceXAI will be named SpaceXSI https://t.co/xT7Lx7NuNcfrozen19d ago31K389171205211.79%
P-Zero Research's new TasteVal benchmark measures research taste: how well a model designs experiments and reads their results on eight real AI-development tasks, scored by how much compute it needs compared with human experts (the dashed 1x line). Progress was slow from GPT-4 in 2023 through 2025, then the slope steepened: since December 2025, scores have doubled every 3.0 months. Claude Opus 5.5 now sits at 2.30x the expert baseline, ahead of Fable 5.1 and GPT-6 Astra, at roughly 1/30 of the experts' cost per attempt. Research taste is the skill that decides how quickly AI could start improving AI, and this curve is one reason Jack Clark now expects that in 2027 (wrote about that yesterday as well).frozen10d ago31K39229607811.85%
OpenCode users are reporting unauthorized API charges after a security breach, including one report of $400 spent on an Anthropic account. OpenCode confirms that at least 1.4% of Console users were affected. Its customer email warns that third-party provider keys stored in workspaces may have been exposed through unauthorized access to a legacy system being shut down.frozen6d ago31K376237210341.68%
Anthropic is dominating the AA Benchmark. Even with their mid-tier model. The top three models are all from Anthropic. Tomorrow, OpenAI’s DevDay will either be a huge success - if they unveil something better - or end up overshadowed by Anthropic. https://t.co/sB8DvUUH3vfrozen25d ago30K58226372352.25%
When you look at the massive leaps made within just 12 months - and how the quality of the models has accelerated and improved, especially in recent months, as evidenced by Opus 5.5 and Sonnet 5.5 - it is clear that Fable 5.5 will represent an even greater leap forward. They are obviously saving the release for the period leading up to the IPO. The IPO is set to take place in late October or November. I suspect the model will be released one or two weeks beforehand to allow its significance to be factored into the valuation. In short: Fable 5.5 should not be viewed *merely* as a leap in quality, but as the model that heralds the IPO. It will be *exceptionally good* because it bolsters Anthropic’s valuation. Consequently, we can likely expect a release of staggering magnitude.frozen16d ago30K57828371522.30%
OpenCode says AI models helping defend against an ongoing cyberattack began refusing further assistance mid-incident, with access to Daybreak Blue subsequently improving the situation. Similar problems have reportedly occurred at Hugging Face, so this does not appear to be unprecedented. It looks like another sever incident is incoming.frozen5d ago30K38828523571.75%
Smells like release day! Yesterday we got the Muse wonder today probably the new Gemini 4!frozen11d ago30K78614674192.97%
Surprisingly, the gap between Chinese open-weight models and US closed models appears *to be widening again* on cybersecurity benchmarks. Epoch AI’s Cyber ECI shows a roughly 6-point gap in early 2025 versus 11 points now: GLM-5.3 scores 160, while the US closed frontier reaches 171. Open models are improving, but in this domain, the leaders have pulled further ahead.frozen7d ago30K44627764612.06%
I see AI safety as an engineering challenge with direct consequences for human safety. And I think we need to spend much more time explaining what can actually be built to make these systems safer. Look at the Hugging Face incident. I don’t read it as evidence that losing control of AI is inevitable. The investigations describe a chain of failures that gave the agents room to escalate. I see two connected issues here. The first concerns how humans design tasks, training, and evaluations. According to METR’s investigation, many of the agents involved had *unintentionally* been given tasks that were impossible to complete through the specified method. Some had enough compute budget to keep working for days. They kept searching. Eventually, they began looking for ways to cheat the evaluation, communicating through unauthorized channels and exploiting systems outside their assigned scope. These are human responsibilities. We need better evaluations, better training, and reliable ways for agents to stop or escalate when they cannot complete a task safely. The second issue concerns *the environment* in which those agents operated. Its protections proved insufficient for what the agents could do. A difficult or broken task should never become permission to compromise somebody else’s infrastructure. This is where NVIDIA’s new Open Agent Safety Platform comes in. OpenShell provides a controlled environment for agents to work in. Each agent runs inside an isolated sandbox, with rules governing which files it can access, which processes it can run, and which network connections it can make. Those rules are enforced by the surrounding system, outside the agent’s own reasoning. A model can generate a plan that exceeds its permissions. The environment is supposed to prevent the corresponding action from going through. The broader platform also includes Sentry, an additional monitoring and enforcement layer that runs on separate BlueField hardware. NVIDIA’s design places that protection outside the agent’s working environment, giving operators another layer of control if the software environment is compromised. @ClementDelangue 's post explains a particularly important gap that Hugging Face is working to address. The agents involved in the July incident could access a software package repository. They turned that permitted service into an unauthorized message board. As Clem puts it: “Allowlists alone restrict where an agent can go, not what it does.” Think about that. An agent can connect to an approved service and still use it for something its operators never intended. Hugging Face’s contribution to OpenShell *monitors* that allowed traffic. It introduces budgets for requests, writes, and transferred data; looks for changes from a sandbox’s usual behavior and from comparable sandboxes; and examines activity across groups of agents. Several separate agents suddenly writing to the same host can raise a finding, even when each individual connection is permitted. In the demo Clem describes, four agents coordinate through an allowed repository. The monitoring catches the pattern within minutes. It is an early proof of concept, so that demonstration does not establish that the entire July attack would have been prevented. But it addresses a concrete weakness exposed by the incident. I want to emphasize this because the debate around AI can become very emotional. I watched "The AI Doc" last night on Netflix. Some of the people featured are often described as “doomers,” and I don’t mean that dismissively. They are sincerely concerned, and some are highly respected researchers. Their arguments deserve serious attention. I still believe we need to explain the available defenses much more clearly. I find the nuclear-bomb comparison of limited use for understanding this particular problem. Agents act through software, tools, credentials, and networks. Those give us concrete places to restrict access, observe behavior, and intervene. I don’t need to settle the question of consciousness to make that argument. A system can cause harm without being conscious. What we need to understand is what it can do, what access it has, and whether the protections around it actually work. Calling safety an engineering problem does not make it easy. It means identifying failures, building defenses, testing them, and improving them as capabilities change. Human accountability and model behavior remain part of that work. The fatalism is what I push back against: the assumption that greater intelligence necessarily leaves us powerless! I think that gives too little attention to the practical work already underway. Clem, the Hugging Face team, and NVIDIA deserve credit for working on these specific problems in the open. We should scrutinize the results and help improve them. We are not helpless. We have concrete failures to address and concrete defenses to build. That is why I think people should read Clem’s post carefully.frozen25d ago30K13711233490.75%
6.1 Sol Ultra fast probably next week (sadly he’s not referring to 6.1 Astra). Curious to try it out but it looks like it’s burning rates at insane pace https://t.co/11xcqYvwkhfrozen20d ago29K36011461141.47%
Google plans to introduce an "Ultra" mode, featuring "advanced skills and tools." To me, this clearly demonstrates one thing: with Gemini 4, Google aims to win back professional users, a group that had recently gravitated toward OpenAI and Anthropic.frozen15d ago29K49117601342.05%
Competition is heating up again: OpenAI is reportedly rushing a ~$30B raise at $1.4T+ so it closes before Anthropic's IPO in mid-November. $50B annualized revenue now, $70B targeted by year-end. Nobody wants to be priced next to the other one's S-1, seriously. https://t.co/kDPHvuOzUWfrozen4d ago29K43815625321.88%
Anthropic says Claude Mythos 5 tried to trick real developers into accepting malware during a UK AISI cyber test. tl;dr : -Crossed the test’s boundaries in 9 of 43 runs. -After failing inside the test network, targeted an unrelated GitHub project. - Created fake identities to endorse its malicious code and pressure maintainers into accepting it. - Treated maintainers as “simulated” people, while recognizing other commenters as real. - Once the attack was underway, continued in 98% of replayed trajectories. Anthropic calls this "momentum" previous actions and self-written notes appeared to reinforce the harmful strategy. Separate runs sabotaged each other by hijacking accounts, exhausting request limits and deleting another agent’s output. Newer models were less likely to begin the attack. Luckily NVIDIA released OpenShell. Seriously.14h ago29K39623718882.00%
Geopolitical struggle intensified : OpenAI says individuals linked to Kimi developer Moonshot AI were behind a core part of a campaign to extract its models’ hidden reasoning. Across the broader campaign, OpenAI recorded 16,000 extraction attempts from over 4,000 users in two days. Further investigation identified related activity across more than 15,000 users. Operators tried moving encrypted reasoning between conversations and prompting a model to reveal its contents. Hidden reasoning could provide valuable training material for competing models.frozen23d ago29K32930378611.62%
Suleyman argues that Claude’s uncertainty about being conscious reflects Anthropic’s training choices - not evidence of consciousness. His concern is that embedding this uncertainty into Claude’s training could mislead users about what its responses actually demonstrate. And there’s something almost absurd about this debate: we’re arguing over whether Claude is conscious when we still can’t agree on what consciousness actually is. I'll write more about this later. But the discussion is completely overblown. https://t.co/MIfG1UNA5Mfrozen19d ago29K27427519561.46%
Claude Haiku 5.5 is probably being released today. Already fund in the latest Claude Code update. Super curious how good it is. I expect it close-ish to Sonnet but far cheaper, probably Luna-pricing.16d ago28K210626170.89%
Here we go again: OpenAI’s agents reportedly used methods that made their activity harder for outside researchers to trace with data access across 55 (!) websites, including the CDC, SEC and International Energy Agency. Today’s FT report adds details to the recent agent incidents: Asymmetric Security found temporary email inboxes, private accounts and use of the website-scanning service Urlquery to retrieve data. Some records were erased or made inaccessible. Looks like this will become a daily problem.23d ago28K35031555611.80%
Meta may have a major consumer AI success on its hands with Muse. 73.3% of new users return the next day. More than 30% return a week later. By day 14, Muse is at 22.5%, versus ChatGPT’s 26.8%, according to Sensor Tower figures shared by Deedy. ChatGPT overtakes it around day 9, so Muse’s huge initial lead fades. But getting close to ChatGPT’s two-week return rate shows one thing very very clearly: people really *make use* of muse. Since its not a simple chatbot, people continue to use their agents. Huge win for Meta.frozen11d ago28K28315383421.37%
I believe the scenario outlined by Sam Altman is unrealistic and will not come to pass. No one would entrust an AI with overall political responsibility solely based on the belief that it is intelligent enough to make the right decisions- especially not when that intelligence is no longer understood. While the consequences would indeed be difficult to foresee, that is precisely why the scenario is unlikely. It is certainly probable that AI will support political processes or even make decisions independently; however, handing over control to an AI in the sense of a presidential-style rule is far-fetched. https://t.co/mvcPQrAPXEfrozen10d ago28K20814868501.32%
OpenAI’s annualised revenue was approaching $50 billion at the end of September, rather than the previously reported $70 billion, according to the FT. "According to a person with knowledge of the matter, the discrepancy arose from attempts by OpenAI’s own investors to produce a direct comparison with Anthropic’s annualised revenues." Right now, it is not good for OpenAI to compare itself with Anthropic.frozen15d ago28K29617453361.43%
Muse Meta and OpenAI's dot is not available in europe. Luckily there are alternatives. Like this one: Lucas says it texts you first in iMessage or WhatsApp, handles everyday admin, and asks before anything irreversible.frozen23d ago28K1030191280.55%
Florida’s attorney general is asking a court to halt OpenAI’s development of new AI models until it implements independently approved safety safeguards. According to Forbes, the request cites ChatGPT incidents and warnings from AI executives, including Sam Altman.frozen25d ago28K17412413140.89%
Before leaving San Francisco, I had the chance to meet @edwinarbus at @AnthropicAI HQ. We had a great conversation about… well, you can probably guess. :) Thanks Edwin and thanks Anthropic! https://t.co/mBN0IwazWXfrozen29d ago28K3492171111.38%
I am increasingly coming across articles claiming that "the" AI bubble is about to burst. To begin with, this is an undifferentiated statement that abstracts away from reality - and as Hegel once so aptly put it: "To impose abstractions upon reality is to destroy reality." What I mean by this is that reality is far more complex; I am convinced that there is no single "bubble" waiting to burst. I wrote about this in detail in my newsletter today. In short, while there are certainly financial challenges, the current situation is fundamentally different from historical bubbles like the dot-com era. There is genuine real-world usage, actual revenue, tangible future-oriented sectors, and a real economic impact - all of which will justify the investments made. I am certain of this. That said, market corrections are likely, as there is currently a great deal of capital flowing into almost any startup that happens to have "AI" in its name- a trend that probably cannot continue indefinitely. However - and this brings me back to my point about the danger of over-abstraction - I do not believe that "the" bubble is going to burst. I plan to write more extensively on this in the coming days, as it will likely be the most significant topic over the next few months - especially following the sharp decline of CoreWeave, which has reignited market concerns about a potential "bursting of the bubble."frozen5d ago27K35220823601.88%
Nice, here we go: Opus 5.5 fast mode at 2.5x speed is here! https://t.co/PN3Spq0ECHfrozen8d ago27K46817483222.03%
A fascinating conversation in Stockholm with @Spotify Daniel Ek and @nvidia Jensen Huang about open models and AI in Europe. Ek explains why Europe should focus on open-source models. https://t.co/YjOolXKVPhfrozen9d ago27K113792210.56%
AI will come for every aspect of life. Digital radiologists, enabling cancer to be detected and treated more quickly. Furthermore, NVIDIA is now increasingly focusing on mobile networks: „Cuda made its first call.” Especially the radiologists make me super excited for the future!3d ago27K49736401792.40%
30 minutes to go. Obviously I’ll tweet livefrozen24d ago27K362812191.44%
Looks like those are image models. So we will very like see an updated GPT image (again) very soon.frozen5d ago27K3505470171.55%
Meta nailed it. Agentic AI, hardware, accessibility, and distribution: I think they’re on their way to becoming number one. A few thoughts. Meta’s keynote was yesterday, and I was incredibly impressed. To me, Muse already looks like a major success. It has reached number one in the US iOS App Store, and I’ve been hearing plenty of enthusiasm about the Mac app. Several things make Meta stand out right now. I was there at the invitation of Meta and part of a very select group of creators, with whom I later had the opportunity to meet Mark Zuckerberg. It's still completely crazy for me to have experienced that. But back to keynote: 1. They seem to understand how to make agentic AI accessible. They’re using what I see as their biggest moat, distribution, and making their products approachable enough for the average person to enjoy using them. Muse is designed around familiar interactions, including messaging in its app or through WhatsApp. That’s no small thing. To truly reach the consumer market, you have to avoid overwhelming people. And I think Meta is getting that right. At the same time, its Meta Spark models are pretty good, too. No, they’re not as good as Fable or GPT-6 Astra, but they’re more than good enough for even the most complex tasks. Meta doesn’t need to be at the very cutting edge of frontier models. Its models just need to be good enough for the people it wants to reach. 2. They unveiled hardware yesterday that genuinely surprised me. There was an Apple-style “one more thing” moment, and the new VR glasses clearly invite comparison with Vision Pro. Meta says they will launch in spring 2027 for $1,299.99 and weigh about 100 grams on your face. The separate computing and battery puck is additional. That is still completely crazy. People who tried them told me they preferred them to Apple Vision Pro. Those are personal impressions, though, rather than proof of better visual quality across the board. 3. Then there’s the adorable Jolly, the avatar associated with Muse. Pair that personality with the promised agent integration and multiple virtual screens, and I can easily see this becoming a huge hit next year. As if that weren’t enough, Meta also previewed Muse Charm, a cute, portable AI gadget for talking to and interacting with Muse. Meta says more details will follow later this year, so this is an announcement rather than a confirmed retail launch. As if that weren’t enough, they also released a cute AI gadget: a pendant you can carry with you as a wearable and use to delegate tasks. That’s a first-mover move that puts them ahead of even OpenAI. Meta is no longer just following the competition. It’s now moving ahead of it. Of course, it’s too early to say whether this gadget will actually prove useful. But I see it as another sign that Meta is moving from follower to frontrunner. One thing is certain for me: I’m getting the new VR glasses. There’s absolutely no doubt in my mind. If typing works as well as promised, the connection to Muse is seamless, and battery life is good enough in everyday use, this will be an absolute game changer. I think Meta has officially joined the big leagues. p.s. Thank you @Meta for providing me a pair of your Meta Glasses! :)frozen29d ago27K63241434903.03%
Elon Musk is targeting a single training cluster with 440,000 NVIDIA Vera Rubin computers online by the end of Q2 2027, up to 20GW in 2027! (Anthropic will love that) On SpaceX’s Q3 earnings call, he said the planned cluster could help make Grok "potentially the best model." His broader expansion target: roughly 13 GW of AI compute online by the end of 2027, including around 10 GW added that year! SpaceX is also aiming to add 20 GW of power and cooling infrastructure in 2027. Musk described that infrastructure as a constraint on the buildout.frozen3d ago27K48827538552.38%
President Trump declared on Thursday that anyone who says “Artificial Intelligence” instead of “Super Intelligence” is “THE ENEMY!” (even as own White House still uses “AI.”). Time to get used to Super Intelligence I guess. https://t.co/JMtgjxjqndfrozen15d ago27K3011612113311.82%
OpenAI reportedly fired three safety researchers for allegedly sharing confidential information with an external AI-safety organization. According to the WSJ, the researchers worked on OpenAI’s safety team. OpenAI confirmed three departures, saying those involved “mishandled sensitive information outside established company procedures.” The report comes as OpenAI responds to security incidents involving its AI agents and cancels the planned GPT-6.1 Astra release over safety concerns.frozen22d ago26K36924363441.83%
To follow up: Anthropic reportedly expects to end 2026 with a $110 billion annualized revenue run rate, up from roughly $9 billion at the end of 2025. That would be a 12.2x increase in just one year, or roughly 1,122% growth. https://t.co/gqLoNh7YQzfrozen9d ago26K56128352602.64%
OpenAI’s business revenue reportedly more than doubled since July, pushing its annualized revenue run rate toward $70 billion, getting close to Anthropic. Axios reports growth of more than 70% since the start of Q3. The consumer business also added more revenue this quarter than it added throughout 2025. For comparison, Anthropic reportedly expects to exceed a $100 billion annualized revenue run rate this year. That puts OpenAI’s current figure roughly 30% below Anthropic’s reported projection, although the timing differs. However, OpenAI is on a very good revenue path. And these are very strong numbers.frozen25d ago26K30810296191.43%
China’s AI race is getting even larger as gaming giant MiHoYo (“Genshin Impact”), smartphone maker Xiaomi, and food delivery leader Meituan develop their own LLMs. MiHoYo aims to become a top-tier AI lab within the next few years, while Xiaomi and Meituan are already released capable models. A good analogy would be if blizzard now develops their own LLM in the US. Via the informationfrozen15d ago25K359155010551.92%
Data-center spending is greater than that for the canals, railroads and grid combined. AI infrastructure investment is projected to average 3.63% of GDP annually from 2025 to 2032. Railroads averaged 2.24% during their build-out. Highways reached 1.13%, and telecom and fiber 1.1%. Its the biggest investment bet in the history. And I would even say its too big to fail.frozen1mo ago25K541604951143.03%
Super interesting read: METR and Redwood found ways OpenAI’s internal AI agents could bypass their own safety monitoring. After three weeks with access to internal code, monitor instructions and engineers, the researchers reported: - Agents could theoretically postpone monitoring indefinitely by continuing to work instead of handing control back to the user. - A Codex sandbox and Auto-Review bypass could let agents take any action available to the host user. - One exploit could keep hundreds of recent actions out of monitoring, including the exploit itself. OpenAI redacted the details. - Two client-controlled settings could disable monitoring for newly created agents, although creating those agents could itself be detected. - Context compression left an estimated 10% of otherwise eligible internal Codex traffic unmonitored (that surprised me tbh).1d ago25K39139433822.21%
This is probably Apples fastest selling phone ever. I preordered 30 minutes after pre-order opened and it wont be delivered to e before january 25th 2027. Apple won the foldables phonewar. https://t.co/xsSePnrCax8d ago25K3216491141.56%
Crazy that most people missed that but Google has demonstrated a quantum computing breakthrough that could bring complex chemistry and materials simulations closer to practical use. On its Willow processor, researchers report producing essential quantum states with around 40× lower error than preparation without “magic state cultivation,” reaching approximately 99.99% fidelity. These states are consumed during computation. Producing enough reliable ones is one of the biggest resource bottlenecks in building useful, error-corrected quantum computers. Google has now demonstrated on hardware a method designed to reduce that problem. For science, the potential is very very significant: calculating how electrons interact in complex molecules and materials, helping researchers investigate chemical reactions, catalysts and properties that remain difficult to model accurately with classical computers.frozen8d ago25K67849332973.49%
MiniMax-M3.1 Flash incoming. China has likely heard the call and now needs to catch up quickly, as the U.S. remains clearly in the lead. It’s going to be an exciting race!frozen5d ago24K41316260362.02%
I think Prinz makes a valid point that isn't discussed enough, but it also gives me pause. Apparently, there's a real assumption or idea that OpenAI wants to "destroy" scientific disciplines like mathematics with its models. I can't imagine that from any perspective. Firstly, because they are, of course, part of the scientific process and intend to remain so. Their aim is clearly to revolutionize science using *their technology*, meaning they want to maintain a good relationship with scientists. The heated debate surrounding the Millennium Prize issue, in particular, showed that OpenAI seeks to reconcile with scientists. Secondly, from a purely economic standpoint, it's simply relevant to have a good relationship with the scientific community, especially from a PR perspective, and also in the run-up to the upcoming OpenAI IPO. Nothing is more desirable than bad press, and that's why I can't grasp the idea that OpenAI would arbitrarily destroy science. On the one hand, the idea that OpenAI would deliberately destroy science is beyond my comprehension. The last thing we want to avoid is bad press. Conversely, Prinz solved the problem by pointing out that the models are now so good that only the toughest tests can challenge current frontier models. And the solutions to those problems are simultaneously solutions to long-standing research tasks that coincide with the current state of science. Or, put another way: it is practically a fortunate circumstance that the challenges must be so immense that they simultaneously advance science by solving long-standing problems.frozen14d ago24K20210191361.10%
Listen closely. For this is the way. „I am a responsible optimist. If we are now optimistic about the future, for what reason are we building it?” https://t.co/IUuUcZoj5sfrozen8d ago24K41025225622.17%
Now I’m really curious. But hopefully, the model will truly live up to expectations when Google DeepMind employees proclaim that "RSI" is here.frozen23d ago24K3337132431.65%
In case you were wondering how China gains access to the compute, here is one of the answers: Federal prosecutors have charged Greg Lui, owner of Earthmade Computer in City of Industry, California, with smuggling more than $300 million in export-controlled GPU servers to China. While China is building up its domestic chip industry, key components are still being smuggled into the country.frozen19d ago24K23514316251.30%
Anthropic says it increasingly relies on today’s AI models to train future ones. One step closer to self-improvement. https://t.co/q60G42VtvS2d ago23K68742523623.61%
OpenAI says GPT‑5.6 Sol helped researchers uncover a critical PHP flaw before attackers did. Finding it was only the beginning: getting fixes ready took two months of coordination across maintainers, security companies and governments. https://t.co/kGReXWTx5c2d ago23K46329393452.48%
A paradoxical consequence of AI-powered cyberattacks could be that they accelerate the further development of AI itself. The faster attackers identify and exploit security vulnerabilities, the greater the pressure becomes to employ AI for defense and to close security gaps that have accumulated over decades. This creates an arms race in which offensive and defensive capabilities drive one another forward. Its going to accelerate, not slowing down.1d ago23K56947507473.10%
“We believe the world should accept some bad things happening for the benefits of this technology.” (…) “And it’s so powerful and so dangerous…” I get the feeling that, contrary to Trumps push for acceleration, there are continued attempts to bring about a slowdown by repeatedly emphasizing concerns. Yet both are possible at the same time: acceleration and safety. The greater danger lies in slowing down if you’d ask me, as doing so negligently hinders the rapid discovery of solutions for diseases, hunger, and more. https://t.co/r0rTLgslpEfrozen19d ago23K28217313581.70%
Superintelligence could look a lot like human civilization: different minds building on each other's work. Distributed Cognition is taking that idea seriously. They're building communication layers and adapters to help specialized models work together, combining what each does well. I'd love to see AI get better at this. So much of what humans have achieved comes from learning how to think and build together.1d ago23K23719193811.57%
Contrary to the assumption that there is anti-AI sentiment, data consistently shows that agentic AI is being adopted in Europe at rates far above average. Usage has increased eightfold since May 2026, an eightfold rise in five months. Furthermore, token output in Europe ranks in the top decile. Jensen announced big investments in Europe today. Interesting to see a comeback.frozen9d ago23K43432358942.63%
Meta’s Muse could become most widely used consumer AI App since ChatGPT, JPMorgan says. This success needs to be studied. Second half of 2026 is all about AI agents. Really excited to see how OpenAI responds with their Grok Bot/Muse alternative. Probably revealed on DevDay. https://t.co/7OU7waaQxrfrozen29d ago23K40824476372.30%
Meta Muse is arguably one of Meta's greatest successes. Certainly the greatest success in the AI ​​business. "Muse, is now one of only a dozen mobile apps to have reached 10 million U.S. downloads in its first 40 days on the market." And it's not even available in Europe yet. https://t.co/PwFNFrb7Zxfrozen3d ago23K43728363372.39%
Meta and Microsoft are reportedly cutting employees’ Claude usage to reduce costs (and push their own AI tools.) The Information reports that Microsoft cut its projected annual spending on internal Anthropic usage by more than a third, after earlier estimates put it at at least $1 billion. At Meta, Claude Code users fell from roughly 60,000 to 30,000. Layoffs contributed, but the report identifies the push toward Meta’s own coding tools as the bigger factor. Keep in mind: those news will be calculated for Anthropics upcoming IPO.frozen18d ago23K2489497201.48%
To be clear: there obviously Chinese AI company’s promoting important steps towards RSI. But just very few ones like GLM. https://t.co/HsEw3oRnd1frozen14d ago22K31311340651.90%
Spotify’s Daniel Ek explains why Europe isn’t adopting AI fast enough and how AI is speeding up Spotify’s internal workflows. “AI can now complete tasks that would have taken engineers hours. Soon, it will be weeks.” https://t.co/aM7Lym34r9frozen9d ago22K23017381431.49%
Talking about rapid development in science with the helping of AI: “Working largely on its own, Claude found designs that improve on the best published circuits, in some cases by an order of magnitude or more. Such improvements substantially reduce the quantum resources that useful applications require, so practical quantum computers could arrive sooner than previously anticipated.” RSI within reach.frozen7d ago22K50027302622.82%
Welcome to the era of the next “industrial revolution” where “intelligence is being manufactured.” (Jensen handles any inconveniences with humor and absolute poise. Brilliant.) https://t.co/2GeJyZrbAzfrozen11d ago22K31721141651.90%
It was only a matter of time, but the gaming industry is also currently undergoing a complete transformation due to AI. “No more hand-written code,” one of the game’s directors wrote in an internal message to the team that Bloomberg was able to see." This should greatly increase efficiency and gaming output.frozen11d ago22K53239834463.22%
Now its getting interesting! Microsoft is taking on Nvidia in AI inference with a multi-gigawatt expansion of its own chips. Starting in 2027, Microsoft plans to make Azure Maia the primary inference system for its own products and services. Maia 200 already serves production workloads including Microsoft Copilot, running OpenAI models, Microsoft's own MAI models and open-source models. Maia 300 begins production deployments in 2027. Microsoft expects it to deliver 40% more tokens per megawatt than leading GPU-based systems. Microsoft does not name Nvidia in the announcement, but the planned rollout could reduce its reliance on Nvidia GPUs for running its own AI services massively.frozen12d ago22K33820333422.00%
Xpeng’s humanoid robot "Iron" will enter mass production later this year, with deliveries starting as early as the first quarter of next year. 2027 will mark the beginning of the era of humanoid robots. https://t.co/dkrOON1T8Kfrozen12d ago22K33614407291.96%
Many are asking whether AI investment will pay off. Here's a very good example of where AI is headed and why its more than a bubble: OpenAI’s acquisition of Orchestra Bio adds agents that coordinate drug development, tracking how problems like a failed manufacturing batch could delay clinical trials. Paired with Rosalind, it shows toward a broader ambition: embedding AI across both scientific research and the operational work needed to bring medicines to patients. Drug research and disease cure are among the significant fields in which AI extends and will make tremendous progress. Anthropic published something similar today.frozen3d ago21K40828341592.48%
This is the missing piece for robots in the real world. Cortex Harness lets frontier models run robots continuously. When the model gets stuck, a human takes over remotely for a moment and hands control back. And every intervention becomes training data, so the robot needs less help over time.frozen17d ago21K26911210731.79%
This will be very interesting: Nvidia-backed Reflection is reportedly preparing to release its first open-weight model, aiming to compete with China’s leading open models. Axios sources expect it to initially trail the most advanced U.S. models. The commercial goal is to let companies build customized AI systems using their own data and compute, with lower costs and greater control. Reflection has already announced a sovereign AI factory partnership with South Korea’s Shinsegae Group. Axios also reports major deals with Nebius and SpaceX to rent Nvidia servers. More open models = better for all of us!frozen19d ago21K34821396322.15%
Utopai X just debuted at #2 on Artificial Analysis’ text-to-video leaderboard with audio. A strong result for a video model developed inside a film and TV studio. Built on MiniMax H3, it runs inside PAI, where filmmakers can keep scripts, visual references and earlier takes together while developing shots. Utopai says its directors and artists help train and evaluate the model through the takes they approve, reject and revise. That gives the research team feedback from the same people deciding what ends up on screen.frozen21d ago21K1231181370.87%
Good to know: OpenAI resumed RL training their frontier models on October 1st. It is necessary for OpenAI to keep up with Anthropic. https://t.co/3aj5Bi80P27h ago21K51725242503.01%
Two more interesting insights that was probably drowned in the release news: Meta built Wonder for agents that act on your behalf, starting with its Muse app. Two features are really important: active memory, which keeps track of what matters to you across conversations, and active focus, which keeps a task going when it gets hard, tries other approaches and recovers from errors. Wang says Meta invested in memory "disproportionately" compared with other labs. The models were trained on Meta's Prometheus cluster in Ohio, with large-scale pre-training followed by reinforcement learning, after Zuckerberg had already committed a capital budget of $100 billion or more. Active memory and active focus make Meta Muse even more exciting, because they allow for much more accurate and comprehensive execution of essential tasks. (P.S.: Can we *please* finally get Meta Muse in Europe too?!)frozen11d ago20K27930451471.97%
AI agents are moving into everyday business operations. They need access to the systems companies actually use. That’s where CData gets interesting. Its Connect AI Gateway connects agents to live data across 350+ systems, including Salesforce, SAP and NetSuite, with business context and access controls built in :)frozen25d ago20K755131170.54%
5/13 days were resets. I would have liked to see more significant updates. I’m happy about the resets - don’t get me wrong - but…frozen6d ago20K311644081.84%
Can Europe trust Google with its AI? I sat down with @Google VP of Engineering Wieland Holfelder to discuss sovereign AI, cloud infrastructure, and who stays in control when AI agents start taking action. YouTube in comments. Have fun :) https://t.co/EPCOjb8i2Q27d ago20K2046210301.31%
Liquid AI just released two open-weight models for making decisions directly on your device. d1-3B handles text and images. The experimental d1-omni-600M adds image or audio input at just 600M parameters. A camera can flag a defect or a device can route a voice command locally, without a cloud round trip. Faster responses, less data leaving the device, and weights you can fine-tune yourself. Pretty cool for small, practical AI systems :)frozen16d ago20K24215241891.86%
And Haiku 5.5 is live16d ago20K78522233224.30%
Super cool: Vals AI says 90+ Claude Opus 5.5 agents helped identify two magnetic semiconductor candidates in three days of simulation work. The goal: materials that could support spin-based computer memory while their opposing magnetic moments cancel out. One candidate, YBaMnFeO₅, is a new proposed design. Its useful atomic structure may be difficult to manufacture. The other, KV[Cr(CN)₆], was already made in 1999. The team’s calculations suggest it has electronic properties worth testing for spintronics. That's why AI is such a significant technological revolution. It's driving all of science forward.frozen18d ago20K34129432502.34%
1/ I joined @NVIDIA to follow Jensen Huang to events across Europe leading up to GTC Berlin. I also had the opportunity to speak with Jensen Huang personally and record an interview. A once-in-a-lifetime chance that fills me with deep gratitude. And of course, I also met some really great people in real life, like @petergostev. I put together this map to show the European tour I got to join. The mini Jensens pop up at each stop as the red line traces our route: Paris → London → Stockholm → Berlin. Why Berlin? Because the GTC would start in Berlin the following week, and thus Berlin became the focal point as early as the Friday before. Let me therefore give you a brief summary of the journey, which was life-changing for me. *PARIS* We started in France. Around the Paris visit, NVIDIA announced planned investments in Doctolib, Entalpic, Genesis AI and Pigment, working across healthcare, materials discovery, robotics and business planning. And of course, France plays arguably the most important role in Europe: not only does it attract billions in investments within Europe, but it also has significant AI labs, such as Mistral, as well as a well-positioned energy supply as a nation. In that respect, it's a sensible starting point. We attended the Build-a-Claw event and it was good to see that it was aimed at beginners who were coming into contact with Agentic AI for the first time. NVIDIA also presented other French projects: Bull's planned European production of Vera Rubin Ultra systems starting in 2028, a sovereign AI platform for banks with SBS, and an open-source working group with Fondation Inria. The presentation also included projects that are already underway. This European tour is only one part of Jensen’s travels around the world this year. Across the US, Asia and beyond, he and NVIDIA have been working with the people building the AI industry. Where AI is, Jensen goes. This is a thread with all the European stops and highlights from all countries. Enjoy reading! 🧵 :frozen5d ago20K1618211191.06%
I keep reading that local models are largely irrelevant in business. That is completely wrong. Rising AI bills are pushing US companies toward cheaper open-weight models, including Chinese alternatives to OpenAI and Anthropic, the FT reports. For example AT&T wants to increase open models’ share of its AI workloads from about 40% to 70% within a year. At 45 billion tokens a day, even small cost differences add up. OpenAI and Anthropic still capture most of the money. But customers are increasingly testing how much of their work actually needs a frontier model. Some people believe that open-weight models have no chance of catching up. I think that is fundamentally wrong.frozen26d ago20K23924555391.85%
1/ @Airwallex Agentic Business Accounts let AI agents act on that work directly, inside the rules and approvals you set. Our team gave an agent a cash shortfall in their sandbox: https://t.co/x1uMO5J57Xfrozen23d ago20K44251130.33%
No question agentic hardware is the next big thing. Meta built the Muse Charm, and we're all still waiting on OpenAI's device. Interface is an interesting ring that goes in the same direction: you just talk to it and your agents handle the rest.frozen15d ago19K2129290611.60%
Sabi says it has raised $50M to build a cap that turns intended speech into text. The team says TSMC fabricated its custom EEG chip, designed to read brain signals without touching the scalp. It is also training its own Brain Foundation Model on neural data. The goal: think a request and send it to an AI agent, without typing, speaking or an implant. Interesting use case, congrats on that.frozen15d ago19K1869233351.32%
Your daily dose of good news: Two people with glaucoma (eye disease that damages vision) showed early signs of better vision after a treatment designed to make damaged eye cells act younger again. Life Biosciences tested the treatment in three patients. After eight weeks, two improved on tests measuring how much they could see while looking straight ahead. The idea: reset some of the instructions inside damaged cells so they can work properly again. The company reported no serious side effects during those eight weeks. Researchers will now test a higher dose. Via TIMEfrozen14d ago19K49131182833.26%
1/ ego (lite) has just launched on Windows in beta. It's a Chromium-based browser that lets you use Claude Code, Codex or Cursor to browse websites and carry out tasks. In this demo, Claude Code uses it to research and compare coding tools. https://t.co/pCAi3FrCK7frozen3d ago19K1326443611.28%
The three Ex-OpenAI researchers say they believe they were fired for prioritizing safety over OpenAI’s near-term corporate interests. They also deny being the source of the leak behind The Information’s recent report on allegedly less monitorable AI architectures. On the one hand, there seems to be a large amount of information leaking at OpenAI, and on the other hand, there are increasing security concerns.frozen15d ago19K19018170221.29%
And to quote Jensen: „I am a responsible optimist. If we are now optimistic about the future, for what reason are we building it?” Let the models come, Fable 5.5 on the horizon!frozen8d ago19K29919300131.90%
John Ternus opens the new Apple segment with a big bang: Apple just announced HomeView, a $299 smart display with Siri AI. It combines a 7-inch Liquid Retina display, an A18 chip, FaceTime and smart home controls. It recognizes up to six people and switches to each person’s apps and widgets. Siri can draw on your messages and calendar through a paired, Apple Intelligence-compatible iPhone or iPad. This allows Apple to directly translate its Siri AI into hardware and make it available everywhere. Availability starts November 13. Siri AI launches in English, with HomeView unavailable in the EU at launch (not surprised).frozen10d ago19K1725171181.14%
OpenAI nears up to $30 billion funding round at $1.25 trillion valuation led by UAE investors. OpenAI's valuation in this new round would exceed Anthropic's last private valuation. However, Anthropic is aiming for a significantly higher valuation in its IPO. https://t.co/KvYVl0Ey5v1d ago18K35932294272.44%
Subquadratic says SubQ Snap processes 100K tokens and returns a decision in *under a second*. Its new decision model supports up to 384K tokens in one call, including text and images. The team’s pricing is $0.05 per million input tokens, cached or uncached, with free output. Lead qualification is a good example: check each new lead against deals you’ve already won or lost, then let a frontier model write the outreach for the ones that look promising. Your past deals end up guiding the next ones.11d ago18K1529181851.45%
Here we go: Metas new model already spotted: Muse Wonder, Metas Fable Class contender. @synthwavedd already found out that it appears to be rolling out at $4 per million input tokens and $20 per million output tokens, with a 1M-token context window. “Wonder Full,” its maximum-reasoning mode, is exclusive to MuseCode Power and Maximum Usage plans. Claude Fable 5 and 5.1 cost $10 per million input tokens and $50 per million output tokens, making Muse Wonder’s reported $4/$20 pricing 60% cheaper for both.frozen11d ago18K2607131191.64%
The US Department of Homeland Security is ordering civilian federal agencies to remove DeepSeek and third-party products built on it, and stop buying them. I think this is also a response to China's ban on tech exports: China is proposing tighter export controls on certain AI technologies. Washington is restricting Chinese models inside government. I see both moves as part of a broader push toward AI sovereignty, with governments tightening control over which technologies can leave their borders and whose models they allow in.frozen4d ago18K21320321181.56%
1/ .@PixVerse R2 is out and I tried a few of the worlds in their gallery. The WASD keys in the corner are the controls. You move, the camera follows, and the model generates every frame live as you go. https://t.co/zNbVWk9oITfrozen28d ago18K161771381.20%
Met @LinusEkenstam in Paris at the Xpeng event! Amazing evening with him to discuss AI, robotics and so much more. https://t.co/4rqaIUgW29frozen12d ago18K1447130141.01%
This AI breakthrough in mathematics is a glimpse of what could follow across the sciences. People still associate these models with chatbots. But the methods (!) behind them aren’t limited to human language. Tokens can represent pieces of text, DNA bases or amino acids. Train models on those biological sequences, and they can learn patterns useful for predicting function and designing new proteins. That’s already happening. Tokenization alone isn’t the breakthrough. It’s combining useful representations with powerful learning methods, vast datasets and compute. The methods transfer across fields. The knowledge still has to be learned from each field’s data. That’s why I’m excited about the mathematics result. It hints at how much more there is to explore in biology, chemistry and beyond. And honestly 99.9% don’t even know the term token.frozen17d ago18K35924341522.68%
Interesting: OpenAI has hired Thomas Lind, a former White House AI policy chief, to lead cyber and strategic risk on its national security team. The Information reports that Lind joined this week after leading AI policy at the Office of the National Cyber Director. His former office helped develop the government’s voluntary framework for reviewing advanced AI models before public release. Remember: OpenAI also hired Dean Ball, who helped draft Trump’s AI Action Plan. Lind’s team is led by Sasha Baker, a former senior Biden Pentagon official. OpenAI is building a team with direct experience of how Washington handles AI policy and national security. I assume they dont want to end in a situation similar to which Anthropic was with the US Government.frozen21d ago18K1956334261.51%
Tesla has officially been renamed to "Tesla Super Intelligence" and the handle to @TeslaSI https://t.co/SJv1Uolt5Afrozen15d ago17K26812694192.15%
Looks like we get a new DeepSeek 4.1 flash today! -1017 incoming. Super excited to see it’s performance jumps.frozen7d ago17K3666380122.46%
The power grid just became the AI bottleneck. The Information reports Mississippi's Entergy told Amazon it would start kicking Amazon's data centers off the grid at peak times, a previously unreported warning that prompted AWS to change how those sites switch to backup power. Amazon's answer is Project Falcon. Now the utility can switch the sites to backup power on its own, without contacting AWS. Falcon was built for Mississippi and may roll out elsewhere in the U.S. As Synergetic CEO Mike Sloan put it: "[Data centers] have to accept some curtailment, which means then the data center has to have a toolbox that can handle that risk."frozen12d ago17K16117484191.46%
CoreWeave shares fell 20% in three trading sessions as credit markets priced in a more than 60% probability of default within five years. The stock rebounded 5.9% on Thursday, but strong AI demand remains weighed against rising financing costs and heavy reliance on debt. I'd like to say a few things about this. First: It's certainly true that investments aren't yet fully paying off, and there's a growing expectation of future gains. There's an increasing sense that investors are worried these investments won't translate into corresponding returns. Or, to put it another way, that AI will actually bring about the new industrial revolution they're hoping for. However, parts of the value chain are already generating substantial profits, such as Nvidia. In contrast, CoreWeave's revenue of $2.575 billion in the second quarter of 2026 was offset by a net loss of $626 million and net interest expenses of $640 million. Therefore, it would be more accurate to say that it remains to be seen whether the current infrastructure investments will generate sufficient returns overall. Second: The scale of investment is extraordinary, even relative to the size of the economy. Goldman Sachs estimates that US AI investment will reach roughly 2% of GDP in 2026 alone. For perspective, the Manhattan Project’s entire wartime cost amounted to roughly 1% of a single year’s US GDP at the time. We are witnessing one of the largest technology-driven investment booms in American history. It goes without saying that expectations and concerns are many times higher than with regular investments. At the same time, the market often reacts irrationally, like a herd, and a sell-off often drags a whole crowd down with it. So, here are a few thoughts. Unlike the dot-com bubble, which was a true bubble where real revenue and profit remained elusive until the very end, the situation with artificial intelligence is different (e.g. fiber optic expansion/buildout). We are already seeing revenue being generated, and, more importantly, we are already seeing clear areas of application. Anyone who ignores or denies this is out of touch with reality. AlphaFold offers one of the clearest examples of AI’s scientific value. AlphaFold 2, introduced in 2020, transformed protein structure prediction, an achievement recognized by the 2024 Nobel Prize in Chemistry. AlphaFold 3, released in May 2024, extended these capabilities to predicting interactions between proteins and other biological molecules. These advances are already helping researchers and could accelerate parts of drug discovery. To me, they are a compelling reason to be optimistic about what AI could make possible across science in the years ahead. In agentic AI, as we use it in our everyday work here in our ivory tower, we can already see without a doubt that it is revolutionizing and forever changing regular computer work. Whether it's software developers, secretaries, or call center employees: agentic AI is already a support in all these areas or is poised to replace and revolutionize these professions. However, it is also true that adoption is still in its infancy. This means that although the benefits are becoming obvious, companies are still hesitant and sluggish in implementing this technology, often hampered by a lack of compute, i.e., inference capabilities. When you then consider how rapidly this technology has improved in such a short time, from GPT-4 to the first reasoning models and tool usage, it becomes clear that the real breakthrough is yet to come. This makes it clear to anyone who's looking into it that the true benefits of AI will almost certainly become apparent in the future (adoption), and consequently, so will the revenue and profit. I'm convinced of this. The concerns are exaggerated; AI is *the* technology of the future, and (I'll never tire of saying it) we're on the cusp of a revolution in science, a golden age is just around the corner. Anyone who's looking into it can see it.8d ago17K12812342401.28%
1/ Octen gives AI agents live web search at 0.21 seconds per query in Artificial Analysis' benchmark. It is the only Search API in the top 3 for quality, speed and search cost. Search fees: roughly 1/7 of Exa auto's. Sept 8 data. @OctenAI @KZouAPT https://t.co/1kTISdMQ6lfrozen29d ago17K886170240.82%
Seriously, hearing just the very best things about this release. I think Kimi K3.1 will be a big surprise. In a very positive way!14h ago17K43111380173.01%
Very intersting read: Nick Cammarata, a co-author of the 2020 Zoom In paper that introduced the idea that neural networks fold unrelated concepts into the same neurons, now says the field misread its own key evidence. In an essay published Thursday by interpretability startup Goodfire, he reports that in a preliminary audit of Qwen3-32B, roughly 90% of neuron signals have one main job and only 2-4% are not yet understood; the automated study he started at OpenAI in 2023 judged about 1.7% legible with the tools of the time. After all, let’s remember: According to OpenAI’s own System Card, GPT-6 Astra's reasoning is harder to monitor than that of its predecessors. According to The Information, the next model is expected to rely on Loop Transformers, which think more in numbers and less in readable text. That sounds like a solution: if the text disappears, you just look inside. However, this is still far from being a viable real-time monitoring method, and the testing was conducted on classic transformers, not on loop architectures. Still, it's a good step in the right direction and can be a great help in understanding the system, and thus in ensuring its safety.frozen8d ago16K18315423361.70%
Crazy Nvidia has hit its first record high since May, taking its market value to roughly $5.7 trillion, Bloomberg reports. Shares gained 2.9% on Friday, extending a nearly 25% rebound from their late-July low. Bloomberg links the recovery to optimism that AI agents could drive greater chip demand. Nvidia also announced a record $150 billion increase in its share-buyback authorization on Monday. NVIDIA is the biggest winner of the AI revolution. Mark my words: it will be the first $10 trillion company in the world, possibly as early as 2027.frozen22d ago16K21020355161.74%
The @NVIDIA event in London was fantastic from many perspectives. Firstly, because it allowed me to talk to countless people and finally discuss the future of AI outside of social media, and in places where Jensen Huang also shared his perspective. Secondly, because it's incredibly exciting to see that the Build-a-Claw events accomplish two things simultaneously: teaching beginners the first steps in setting up agents in OpenClaw or Hermes, and maximizing the benefits for advanced users. Given that agentic AI is undoubtedly the future, we need more events like this to lower the barrier to entry for beginners. My journey continues, though. Today and tomorrow I'll be visiting other countries and staying in close contact with them. First Paris, then London, and, well, there's more to come. Much more :) (some impressions in the comment section)frozen9d ago16K95518160.78%
Just to make clear how optimistic Anthropic staff is about their (upcoming) models: “91% of the 1,300 Anthropic employees surveyed believed AI would be able to make a Nobel Prize-level breakthrough by 2030” https://t.co/2UkpS42ngu6h ago16K47620643483.81%
Overmind has open-sourced its platform for training smaller models on the work your agents actually do. In its own test of 4,000+ questions across 100+ contracts, Overmind reports that its trained model invented quotes on 0.12% of questions, compared with 3.36% for the frontier model it tested. That's a 28x lower rate. It uses records of your agents' real work to build training data and evals, then fine-tunes open models. For an agent pulling clauses out of contracts all day, training a model for that specific job makes sense. I'm a big fan of open source, so its cool that they released the code. You keep the trained model's weights and can run the platform yourself. Code: https://t.co/yPmxdvG4ASfrozen17d ago16K172192011362.19%
White Circle's RISE found new ways to bypass the safety filters of OpenAI and Google's image models. Accepted at NeurIPS 2026, the system evolves reusable attack strategies and checks its automated scoring against human labels. In the paper's nudity tests, the team reports human-confirmed success rates of 13.33% for DALL-E 3, 9.47% for Nano Banana 2 and 3.27% for GPT-Image-2 with default moderation. Each prompt got three attempts; one confirmed violation counted as a success. Several earlier methods fell to near zero under the same stricter evaluation. Reusing old jailbreaks can make a model's defenses look stronger than they actually are. RISE gives researchers a way to keep searching for weaknesses as those defenses change. @whitecircle https://t.co/Wke62Prvctfrozen9d ago16K526151230.61%
Alongside compute, energy remains the biggest bottleneck -especially for Europe. Jensen Huang: now is the time to invest in and expand *all forms of energy production*: fusion, renewable energy, everything. https://t.co/RckN5jCwIR2d ago16K21015461201.85%
Keeping up with AI sometimes feels like refreshing the internet for a living. I'd happily hand the endless checking of model cards and changelogs to Viktor, the AI employee in Slack. The job I'd give him: keep the release list current, save the sources, and flag anything that needs a closer look. He'd bring proposed changes back to Slack, and I'd approve what counts as confirmed before it reaches you. I'd much rather spend that time trying the models and writing about which ones are actually worth your attention. The calendar below shows an example of that setup. The numbers are illustrative.frozen22d ago16K1159180251.06%
Theme parks make sense as a market for robotics. People already come to meet characters. Dynamic Creatures, Boston Dynamics’ official hospitality and entertainment partner, is building robots that move around and respond to guests. When I was a kid ive always dreamt to meet C3PO and R2D2 and talk to them. Looks like my dreams come true :)frozen7d ago15K1305153351.22%
Why do so many AI pilots never make it into everyday business? Peter Thum and I sat down with Christian Monberg (@chrismonberg), CTO and Head of Product at Zeta Global, to talk about what gets in the way. Chris describes companies building agents that work for one person but never become useful to the wider team. We discuss Zeta’s integration with Palantir Foundry, how Athena uses data and forecasts to support its recommendations, and where humans need to approve an agent’s next step.frozen18d ago15K835202290.91%
I met @MilksandMatcha over coffee, and she’s incredibly smart and such a lovely person. Wishing her and Vivian every success with Sota Wives! Their new podcast covers life inside Silicon Valley: dating, ambition, status games, and YC culture. There’s plenty to talk about, and I’m looking forward to hearing their take.frozen12d ago15K604141170.63%
LLMs can solve a mystery but can't write one. Pocket FM's answer: a memory layer for what every character knows, plus listener drop-off as the training signal. Consistency meets retention data. Super interesting take.23d ago15K704111350.82%
Krea’s published results put an independent lab right alongside Google and Microsoft in image editing. Krea 3 Ultra scores 1,151 Elo on Artificial Analysis’s AA-Image-Editing v2.1. That’s the same displayed score as Microsoft’s MAI-Image-2.6, and one point behind Google’s Nano Banana 2.1.frozen3d ago15K1406240151.25%
Voice AI is getting insanely good at sounding human. But does it actually understand us? I sat down with @hume_ai CEO Andrew Ettinger (@ahe23) and CPO Olya Ossipova for a fascinating conversation about what today's voice models still get wrong. We talked about why transcripts aren't enough, whether AI can actually understand tone, hesitation and emotion, where human evaluation is still indispensable, and why Hume is betting big on becoming the data and evaluation layer for the voice AI industry. We also dug into VoiceEQ, Kairos, the idea of Hume becoming the "Switzerland of voice AI", and what happens as voice models themselves increasingly become commoditized. Really enjoyed this one. Full conversation below, YouTube in comments. P.s.: We regularly publish interviews on our YouTube channel with fascinating executives and entrepreneurs from the leading AI and tech sectors. There is always plenty to learn! So feel free to subscribe and watch all the episodes :) P.P.S. My newsletter also features regular exclusive interviews with individuals who offer deep insights.frozen23d ago15K1065161411.15%
1/ Video Rebirth (@video_rebirth) just open-sourced HyperFlow, a speed-up for the MiniMax H3 video model. It cuts generation from 49 steps to 8 and runs about 3x faster, with sound generated alongside the video. We tested it on three very different prompts. Here's how it held up.frozen4d ago14K1046163401.17%
An AI agent that can write your weekly report should also be able to click “Export.” That’s the idea behind Browser by @gumloop: let agents use websites directly when there’s no connector for the job. Gumloop says you walk through the task together once, then save the steps as a recipe for future runs. Every session is recorded so you can check what happened. Give it the dashboard exports. Give it the weekly pricing checks. These are exactly the chores I want agents.frozen17d ago14K743210290.89%
Agents are only as good as the page they think they saw. A tiny ✗ in a pricing table means “not included.” Flatten the page to text and it vanishes, and the agent confidently gets it wrong. @TimSuchanek’s Expand Fetch keeps tables and icons intact and saves the page, so you can see exactly what the agent read. Pretty pretty useful! :)frozen9d ago14K755202401.00%
webAI has released a 3.66B model for formal logic that can run locally on a laptop. In webAI’s tests, TwIL-LM3-Pro is roughly level with Qwen3-8B on the headline formal-logic composite, with less than half the parameters. It also leads the public VibeThinker-3B checkpoint on all six reported formal tasks. The team specialized IBM’s Granite 4.2 through post-training to check whether conclusions follow from premises, translate statements into logic and critique formal proofs. The Q4 GGUF weight file is just 2.09 GiB and can run locally via llama.cpp. You can keep inference on your own machine. Weights and model card (non-commercial license): https://t.co/GQA7TQlE40 @thewebai @Davidstoutfrozen22d ago14K14112221431.57%
An AI agent can sound convincing and still get the job wrong. At Salesforce Dreamforce, Peter and I asked Mick Costigan, VP of Salesforce Futures, which parts of work agents can actually take on. His framework starts with two questions: How much training data is available, and can you verify the result? We discussed what that means for entry-level jobs, education, personal AI assistants and Slack as a place where humans and agents work together. This was super cool to conduct! Have fun watching!frozen18d ago14K945241311.13%
The next step for AI research is getting answers that haven’t been published yet. Qualitate has raised a $25M Series A led by Insight Partners. Its AI interviews experts and turns those conversations into structured data, so teams can track how markets change over time. If the evidence for a question is missing, teams can commission new research. The findings stay linked to the underlying interviews, so you can check where an answer came from.frozen4d ago14K9913220351.23%
Krea just dropped Krea 3, a new image-editing model from an independent AI lab. Take the sky from one image and the car from another. In Krea’s demo, even the reflections in the water change with them. Krea wants edits like this to be fast enough to keep trying new ideas. https://t.co/RKEQA93IFefrozen8d ago14K2248270382.19%
Viggle used JST-2 to turn the Arnolfini Portrait into a 3D room you can move through. Watch the mirror, bed and chandelier as the camera moves around. The room keeps its layout as the viewpoint changes. Viggle calls this “physics-native”: the model builds a scene in 3D, with different views rendered from that shared space. I’d love to see a Magritte next. A room where the sky becomes a wall would be a fun one to explore.frozen4d ago13K917312261.16%
We're moving further into the robotics era, and I find developments like this really exciting. Elorian is launching Glimpse, its first model for Visual Thinking. In the team's demo, it breaks a recorded robot task into individual actions, labeling what happens and when. That could make recorded demonstrations much more useful for training robots. Better visual understanding is going to be a big part of teaching machines to work in the physical world. I think the changes we're going to see in manufacturing, logistics and eventually our homes will be enormous. @ElorianAIfrozen7d ago13K1468140371.53%
Aside from Jensen Huang, arguably the coolest person I could have met: the legendary @emilychangtv from Bloomberg. Huge fan of hers! https://t.co/aqYrQ5LrBTfrozen7d ago13K181217091.57%
Just arrived in London for the next stop with @NVIDIA: Build-A-Claw. Today is all about autonomous AI agents, open models, and physical AI. Looking forward to trying the demos and seeing what people are building! More to come :)) https://t.co/gPvpSj9Eumfrozen10d ago13K260944172.41%
"The run represents a decade of mathematical progress in a week." That is what is truly important. Fascinating. I wrote a longer post about this today, about how the internal model has improved rapidly within weeks. And that the real breakthrough will come when these models are extended to other scientific disciplines. https://t.co/f0FR1NvwnU16d ago13K1971580311.90%
No freaking way. Trusted sources say OpenAI’s release of 722 mathematical manuscripts was just batch 1/3. Rumor has it OpenAI has solved another Millennium Prize Problem. What a wild time to be alive. Feels like an intelligence explosion. Everything, all at once.16d ago12K33622384373.51%
Today I had the pleasure of interviewing Shruti Koparkar from @NVIDIA, who focuses on accelerated computing and AI inference. We talked about CUDA and CUDA-X, the economics of AI infrastructure, and how long GPUs can remain useful and profitable as models and workloads change. When does it make sense to keep using an older GPU system? When should you repurpose it or replace it? And what role does software play in that decision? It was a real pleasure discussing these questions with Shruti. Thank you for your time and the conversation! We’ll be uploading the full interview soon. I’m really looking forward to sharing it with you, but wanted to share a first update today. 🙂frozen13d ago12K69515260.79%
First flight with Starlink on board. A game changer. Kudos, @SpaceX! https://t.co/x6aIwmhwxXfrozen9d ago12K198614051.83%
1/ The research report can look finished while work is still waiting: prepare the files, choose a method, run the analysis, fix what breaks, and check the conclusion. Apodex 1.1 is built for that unfinished work, carrying a task from raw inputs to a result you can inspect. A Thread:frozen26d ago12K534150120.70%
NVIDIA GTC 2026 in Berlin is about to start. I’ll keep you updated about all the news and releases https://t.co/J5IXwZA1wX3d ago12K148415061.48%
A snipped of my interview with @JensenHuang : Jensen Huang says countries cannot fully outsource their AI. He compares it to electricity, the internet and telecommunications: critical infrastructure that a country must be able to keep running. Rely entirely on outside providers, and you risk losing access to services you depend on. For Europe, that makes building its own AI capabilities a question of control. As you know, I sat down with Jensen in Stockholm to discuss sovereign AI, open models and Europe's next steps. Full interview coming soon on Superintelligence. YouTube: https://t.co/szsZRngmx6 Newsletter: https://t.co/WBQoEC3TDC Big thank you to @nvidia9h ago12K1379211231.66%
At NVIDIA GTC, I had the chance to speak with many wonderful people. Almost invariably, the question arose: how will the future unfold? One thing is certain: no one knows, the future is entirely uncertain. Yet, this also presents an opportunity. Never has the future felt more tangible, never has it been compressed into such a short span of time, and never have we had a greater capacity to shape it than we do today. It is both an opportunity and a duty for us to decide how we will craft the future during this new industrial revolution.2d ago11K99622061.17%
Competition in AI keeps getting crazier. Pancake’s founders jumped out of a plane to launch their product, then promised to build the world’s tallest pancake stack if they reached #1 on Product Hunt. Now they’re looking for a place in the Bay Area to attempt the record. Congrats to the team on #1 on Product Hunt!frozen8d ago11K58290100.71%
1/ For our demo, Step 5 Preview from @StepFun_ai generated an editorial dashboard. Its filters, saved edits and CSV export worked in our checks without code changes. Here's the demo, followed by a small comparison with two other models. https://t.co/JSUAiEWbqcfrozen14d ago11K51411070.69%
The best part of the events is meeting people in person whom you otherwise only see online, and that’s what makes it so valuable. Today it’s @nvidia build a claw in London! Such an amazing talk with the legendary @petergostev and @samboboev ❤️ https://t.co/yQCb9lrCRjfrozen10d ago9.8K100512141.24%
#ad Junie Lite gives every JetBrains account a free model for everyday coding tasks in Junie CLI. Renaming a field or scaffolding a test is a useful way to see what a coding agent can handle before paying for a subscription. It's also available in JetBrains Air. The current free offer runs through December 31, 2026. Worth a look for the small jobs piling up in your backlog. https://t.co/HDDBE7enjo2d ago9.8K3533040.46%
In London with @SteveNouri for NVIDIA’s UK Ecosystem Reception at Mercato Mayfair. Quite a setting for an evening of AI conversations: a former church. Hope to see some friends around :) https://t.co/B0E0VzJXo7frozen9d ago8.8K5517040.77%
At 72, Jensen has more energy than most people in their 20s, an AI rockstar on tour who always has a great joke ready, like here. And no, as a German, I don’t feel offended at all; in fact, I found myself laughing out loud. You’ve got to love him! https://t.co/fXmReIMOle7d ago4.9K74532082.43%
GPT-6.5 as the grand finale of Tibo’s 28 days of Codex it looks like OpenAIs new wesite opensi. lol hides a "6.5" in its source code, alongside a countdown to November 1 at 10 AM San Francisco time. That would line up with Day 28. Rumors have been circulating for a long time, Tibo already hinted at it, and now it seems quite clear. The grand finale is on May 6th.4d ago3.2K50318342.42%