← Back to leaderboard
cgtwts

CG

Verifiedactive
@cgtwts

startups, shitposting & building.

RosterAI Model News & ReleasesHumor & Meme Accounts
26,889 followers · 985 following · Last polled 3h ago
Perf. Score
61.7
Rank #129
Value Score
38.7
#170 of 273 priced
Median views
4.9K
p25 2.0K · trailing 7d
Followers
27K
+5.2% 7d
ER (impr.)
2.39%
eng ÷ impressions
ER (foll.)
3.88%
eng ÷ followers
Est. CPM
$61.43
QT-rate basis
WoW views
-19.7%
flat

Rates & value · est. from 4.9K median organic views

overpriced
Quote tweetbasis
$300
$61.43 / 1K views
Post
—
no CPM
Thread
—
no CPM
Retweet
—
no CPM
Value Score 38.7 · rank #170 of 273 priced$2830.19 / 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 (4.9K), not the mean (44K) — so a single viral post doesn't inflate a creator's reach. The p25 floor (2.0K) is the level 75% of posts clear.

Recent posts · ranked by views

PostPostedViewsLikesRepostsRepliesQuotesBookmarksER
“Sir, tavus just built an AI that can pass the turing test on a video call by seeing, hearing and reacting like a real human in real time” it’s over for us. https://t.co/WBxP18p363frozen21d ago2.0M8.9K381116223.5K0.63%
anthropic’s lead engineer btw. https://t.co/pTlsbqDlXmfrozen8d ago2.0M19K844722834K2.66%
holy shit, this is literally the stimulation hypothesis.
frozen
5d ago
1.7M
44K
1.8K
192
53
7.5K
3.04%
All those years of refusing to learn excel and PowerPoint is about to pay off https://t.co/3L5XJRELNFfrozen16d ago1.6M49K3.8K1821465.4K3.69%
grad students in the EU watching OpenAI watermark AI text: https://t.co/zptKSvWtzqfrozen17d ago1.1M10K22786161.5K1.07%
mark zuckerberg might be the closest thing we have to steve jobs in this erafrozen29d ago860K1.1K5722366760.18%
RIP Git https://t.co/uIeRj2Qi3S9d ago837K4.6K164102311.7K0.79%
this is literally steve jobs coded https://t.co/AeF2ElrBpzfrozen29d ago835K10K4522511502.6K1.67%
Anthropic employee btw.frozen11d ago689K2.5K303321.4K0.58%
silicon valley literally had everything https://t.co/O7MtxupEQffrozen5d ago562K6.9K39361672.1K1.69%
Dude. https://t.co/GPpRG0BrHa9d ago461K1.7K301923640.46%
RIP internet, it was fun while it lasted. https://t.co/bd5vb9fACM15h ago403K2.8K337113501.4K1.17%
Dude. https://t.co/m3WUKbsB2q10d ago242K14K1.0K4597026.44%
“Sir, they’ve removed ‘em dashes’ from Opus 5.5. it’s so over for us.” https://t.co/K0SLJINwFFfrozen1mo ago238K3.5K802061981.58%
please don't nerf Opus 5.5 please don't nerf Opus 5.5 please don't nerf Opus 5.5frozen26d ago223K5.8K1287791992.76%
math is solved it’s so overfrozen16d ago204K1.3K393041540.77%
Dude. https://t.co/JgSEIyYCNzfrozen11d ago188K2.3K374554531.52%
Holy shit, they just killed notion.frozen23d ago175K1.1K322823950.89%
wait, how does anthropic have more compute than openaifrozen1mo ago174K1.9K188231181.23%
internet lockdown soon?! https://t.co/0qQori1ibr7h ago156K1.5K11446153671.32%
physics is solved. it’s over. https://t.co/ngVxHhj5DAfrozen8d ago149K1.1K8355135191.17%
Dude. https://t.co/jiCjXnPTKK2d ago147K2.9K6454283202.26%
“who should we lay off first” CEOs: https://t.co/4HAFAlHPNY10d ago142K2.8K14022112612.25%
Anthropic researcher btw.frozen1mo ago123K2.5K34601442.18%
muse charm might be meta’s best product in the last decade https://t.co/DYkknx44rbfrozen29d ago115K6193840191560.76%
scenes at Railway HQ right now: https://t.co/QBrLhKCQeK22h ago109K2.0K1321873372.24%
scenes at OpenCode HQ right now: https://t.co/VCpG1hGfmhfrozen5d ago104K1.9K1231482742.25%
Silicon Valley literally predicted everything. https://t.co/0PdyAfKno99d ago93K1.2K86552551.72%
holy shit. someone built an app that reads thousands of X posts and reddit threads and tells you what people actually think asked it the oldest debate in football: messi or ronaldo? 64% messi 28% ronaldo 3,419 opinions. it's not even close 💀 https://t.co/NJJkeeU6Ff10d ago84K1.3K783331.6K3.61%
Anthropic right now; https://t.co/CGFxTWIYZefrozen24d ago84K1.9K19162402.32%
“Sir, they just dropped a decision model that handles 384k tokens, makes decisions in under a second, scores 99.8% on long-context tests and charges $0.05 per million input tokens with free output tokens” it’s over for us. https://t.co/dRxP7MzRBw10d ago71K692107703621.63%
the lead gen industry is cooked zoominfo charges $600/mo. lusha charges $449. treg charges $0 in subscriptions and just $0.0089 per lead. the pricing-to-value ratio is actually insane. https://t.co/9ioc6x5QtB10d ago69K538291746011.71%
they’re solving history now it’s so over. https://t.co/xm40sESRY5frozen21d ago67K34415511500.76%
google either went insane with the benchmaxxing or dropped a generational model. there’s no in between.frozen22d ago64K57514395471.06%
qoder just added qwen3.8-flash you can use it free until sept 30 with no claim or paid plan, and it won’t use any credits during the offer. if you’ve been curious about the model, this is a pretty easy way to try it without worrying about burning through credits.frozen25d ago64K1912050.04%
anthropic just solved everything we are so back. https://t.co/ZS1CRBsLCZfrozen29d ago62K3081432820.66%
ElevenLabs founders right now: https://t.co/BqJm9mD4VE9d ago61K35644721370.89%
Claude is down btw. https://t.co/FI0v0oOwaAfrozen23d ago57K1.2K28172382.35%
motion designers right now: https://t.co/uTbXrABLHbfrozen14d ago56K49023931181.15%
anthropic is so back “You should Know”.frozen20d ago55K2749201920.87%
“we’d be profitable if we didn’t lose $80 billion” https://t.co/hS2JcbSg9Q2d ago54K6.1K26241911012.12%
Higgsfield just hit $1B ARR in 18 months > be higgsfield > launch an AI creative suite 18 months ago > bet on the application layer > keep shipping > 80 releases and feature updates since June > Fortune 500 companies start adopting it > enterprise adoption grows 10x > pass 30M users worldwide > go from $100M to $1B ARR in 10 months > with 400 peoplefrozen28d ago50K336136155791.14%
RIP openai. https://t.co/17QIvgu5vy3d ago49K100480250.28%
they just killed like 100 startupsfrozen16d ago48K2648130720.75%
anthropic’s comeback needs to be studied. it’s so over for openai.frozen13d ago45K45891301931.51%
holy shit, this looks bad. https://t.co/OX4TbNvyTkfrozen21d ago39K73530320.29%
anthropic is so backfrozen21d ago37K134350570.54%
they just killed grok botfrozen23d ago34K294101032241.27%
claude just killed 10000 startupsfrozen14d ago34K1907131890.89%
this keeps getting worse. https://t.co/1UxTG6oXpMfrozen21d ago31K125340140.48%
anthropic is on an absolutely legendary run right now3d ago30K3951060521.56%
Anthropic CEO after realizing DeepSeek is almost matching Opus at 1/10th the price: https://t.co/5a6KcX1xpufrozen6d ago29K42032135501.82%
ex openai researcher btw.7h ago29K3482050811.59%
the person who spent 6 years at apple working on apple watch is now building an ai agent that predicts migraine flare-ups and handles healthcare from start to finish. it connects wearables, medical records, labs and symptoms, then takes care of doctor appointments, prescriptions, insurance and medication delivery. this is the future of healthcare.2d ago27K1687030150.93%
Holy shit, an ex-Stripe and Uber engineer just solved a problem every company building AI agents is going to run into. falconer keeps company knowledge and documentation in sync with the code, so agents aren’t making decisions based on outdated information. the next big bottleneck for AI agents is context, not intelligence.1d ago26K941031490.60%
🚨 BREAKING: Fable 5.5 will launch in the next 12 hours3d ago25K60615487182.74%
holy shit, they just killed grok bot, muse, dots and openclaw.1d ago25K1385211490.85%
anthropic is literally insane.frozen19d ago25K70340190.39%
1,000 browser tasks cost: Drex: $2.80 Claude Code: $31 Codex: $25 didn’t expect the cost gap to be this big. small models might end up being a huge deal for running agents at scale.2d ago24K10832160390.80%
Chinese AI labs watching a model outperform larger models on coding benchmarks while casually running at 125 tokens/sec on a MacBook Pro: https://t.co/PX30oCSlhJ1d ago24K1693150401.03%
“it's been 10 weeks of 50% week-over-week growth” it’s so over.1d ago23K60720360.45%
DeepSeek is the real openai btw.2d ago22K8073471374.08%
holy shit, AGI might be a lot closer than people think. AI is now breaking software on purpose, finding bugs and fixing them before a single user touches it.1d ago21K1351030500.92%
as AI agents get better, the amount of inference they need is going to explode. making all those model calls faster and cheaper is a huge part of the puzzle. interesting to see General Compute and Positron working on this.2d ago21K56103070.36%
holy shit, they just killed Jev.frozen16d ago21K84370310.59%
they just solved one of the most annoying problems with ai coding agents, forgetting everything you taught them. and they open sourced it 🤯2d ago20K11711801041.19%
the cybersecurity industry has had a generational disaster of a month. they’re calling it black october for a reason. > Oct 5: SubQuery npm package backdoored > Oct 7: IDCF Cloud hit by ransomware, affecting 495 organizations > Oct 8: TensorLake publishes malicious npm package > Oct 14–15: Zeabur infrastructure breached > Oct 15: GhostAction targets GitHub repos with malicious workflows > Oct 15–16: Rank Math PRO update delivers malware > Oct 16: ConnectWise cloud backup identities compromised > Oct 17–18: TeamUpdraft customer data compromised > Oct 18–19: OpenCode exposes AI API keys > Oct 19: Nikkei reports 1.83M records exposed > October 19: OVHcloud CLI binaries reportedly backdoored > Oct 20–21: PHP critical RCE vulnerabilities disclosed > Oct 21: LiteLLM ships malicious PyPI + Docker releases > Oct 21: Railway reports potential exposure of customer secrets > Oct 21: eMudhra certificate authority keys compromised > Oct 21: BIND 9 security vulnerabilities disclosed > October 21: Node.js security vulnerabilities patched > Oct 22: Linux kernel vulnerabilities accidentally posted publicly, severity up to CVSS 9.0 > Oct 22: Capcom reports unauthorized access affecting 11.77M records malicious packages. stolen API keys. backdoored updates. cloud breaches. critical RCEs. compromised certificate authorities. and we still have over a week left in October 🤯8h ago20K299532361642.72%
Dude. https://t.co/DASlxBkVNjfrozen24d ago18K3163160101.89%
holy shit, you can now ask claude or chatgpt to build an app for literally any tiny problem then share it with your friends like a google doc this changes everything.15d ago18K101290480.90%
they just killed like 1000 cloud-based agent startups.frozen29d ago18K134471270.98%
holy shit, they just killed gold.16h ago18K1761151491.38%
Dario: we need to slow down AI development Anthropic a week later: introducing Claude Opus 5.5 Anthropic another week later: introducing Claude Sonnet 5.5 https://t.co/gsVYISMubEfrozen24d ago17K75130173304.80%
holy shit, imagine selling your car to make an AI-generated GTA VI movie this is either the future of filmmaking or the worst financial decision of the year 😭frozen4d ago17K681270190.63%
they’re about to kill like 2,000 startups tomorrowfrozen24d ago16K93470210.79%
startups right now: https://t.co/j15PNuDdWgfrozen16d ago15K114860431.13%
so they just solved a huge problem with testing AI agents. Era by Eon gives agents a realistic enterprise environment where Salesforce, Slack, Zendesk, Jira and more are all connected, with the messy data and real API behavior agents face in production. You can reset the company, change the model or prompt, and rerun the same tasks. and they’re giving it away for free.frozen17d ago15K46126080.48%
this feels like the GPT-2 moment for world models https://t.co/44eqqPnT44frozen28d ago15K60440240.61%
i think browser use is still the biggest problem for AI agents like dots, grokbot, muse and others. Gumloop’s new Browser might be the missing piece. agents can actually use websites, save workflows as recipes, and run everything in the cloud without you babysitting them. https://t.co/qLwwboJDKmfrozen16d ago15K721790200.78%
>PHP is having a security nightmare tonight >Anthropic drops AI that catches vulnerabilities before hackers do you literally cannot make this up 😭 https://t.co/sScHeVnbRY1d ago15K109650291.01%
the .si domain meta is about to get insanefrozen18d ago14K1475192201.39%
Ben Affleck is incredibly smart bro went from making movies to understanding tensors, CNNs, GPUs and transformers, getting private demos from Google and OpenAI, then building an AI video company that Netflix bought for $590m https://t.co/Rk9hG7pU9dfrozen16d ago14K19113113642.06%
marketing in 2026 is going to get absolutely insane. Higgsfield Content Factory lets a solo founder create more ad variations than an entire marketing team could dream up. we’re still so early. https://t.co/iEpBp1Tzzn10h ago14K681120731.13%
Dude. https://t.co/wNwCXyd5Sffrozen22d ago14K4623060.42%
Holy shit, 1M token output limit is literally insane. google is so back. https://t.co/w707HUPFwKfrozen22d ago13K1981640171.74%
openai must be SHAKING right now… “We found DeepSeek V4.1 Flash 1017 beats GPT 6 Sol on our multi-step offensive cybersecurity task” https://t.co/a0Vp1W2YKw2d ago13K101791281.10%
they just killed like 10000 startups8d ago13K91440270.95%
they just solved one of the biggest legal headaches for startups no more waiting days for contract reviews or paying unpredictable legal fees here’s how their AI-powered law firm works: > connect Slack > send over a contract > approve the flat fee upfront > AI + real attorneys handle the work > get your reviewed contract back in 4–6 hours legal was long overdue for this.frozen21d ago13K47770110.55%
“the limit isn’t how fast we can communicate to machines, but how much they can understand without being told.” imagine a world where you don’t type prompts anymore. you just think, and AI gets to work. we’re getting closer than people realize. https://t.co/N2LK7k6xFbfrozen5d ago13K1052040251.19%
insane opportunity for founders just comment your website on the original post and Relay will build your entire content library for free in under 30 seconds 🤯frozen21d ago13K34230310.55%
general compute teaming up with cerebras to push toward 750 tok/s is actually crazy. GPUs handle prefill while cerebras handles decode, letting each chip do what it does best. this could be a big step toward making ultra-fast inference available at scale.frozen23d ago13K113152071.08%
they just killed google geminifrozen16d ago13K153760261.53%
> open https://t.co/sQU63VNJjh > it redirects to the Grok app download page SpaceXAI really bought it right after OpenAI announced Dots 😭frozen23d ago13K1487175101.49%
Mark Zuckerberg: “It is a BIG watermelon…” meta is so back. https://t.co/nC6ZoF05Iefrozen10d ago12K72240150.74%
so they just solved the problem of AI agents checking AI-written code ContextQA launched Ship, an Autonomous Quality Engineer that tests software, reproduces bugs, investigates failures, and hands the context straight to Claude Code or Codex. https://t.co/FobpwEtKdB17d ago12K7411140271.05%
they just killed like 1000 startups10d ago12K155580581.89%
“Sir, ex-DeepMind researchers just beat GPT-6 Astra on 4 visual benchmarks at 6% of the cost” https://t.co/F2nvyWZvZvfrozen6d ago12K77940110.85%
holy shit. this might be one of the biggest opportunities in AI. the US healthcare industry spends over $1 trillion a year on staffing and IT. imagine how much of that AI agents could replace. has this company figured it out? https://t.co/F9EgwRzplb8d ago12K841440291.13%
holy shit. https://t.co/BJpFUFEukDfrozen13d ago12K6973090.76%
“I guess Dot’s having a slow morning”. this has to be one of openai’s most generational aura fumbles ever https://t.co/wzlSNXzx79frozen23d ago11K119362201.31%
everyone’s talking about AI taking jobs, but what about the jobs AI is creating? the automation engineer is a perfect example. building automations is getting easier by the day, but knowing what to automate inside a real company is a whole different game. people who understand how businesses work and know how to build with AI could have a massive opportunity here.frozen21d ago11K6413100230.96%
“Now… with better WiFi”. https://t.co/VzSVBOTDmhfrozen21d ago11K43340100.54%
So they just solved the biggest problem with personal agents, they can reach you through messages instead of another app or browser, with Photon as the infra layer underneath. grok bot, muse, dots and more are already showing us where this is goingfrozen21d ago11K60980210.88%
i still think about this https://t.co/56mQA4Tgq0frozen11d ago11K25513163483.06%
> have an idea > build an app with AI > form an LLC > open a business bank account starting a company is becoming a one-click process.frozen6d ago11K93640411.34%
holy shit, this is way better than a reset. tibo and openai are so back.frozen17d ago11K1364100101.51%
pov: you’re watching the singularity happen in real time “After 21 hours spent searching this data by roughly 950 agents using 210 million tokens, one of the agents spotted something remarkable: a repeating pattern of DNA sequences that occurs next to the gene for an odd-looking RT.”frozen29d ago9.9K918162311.50%
SaaS CEOs waking up to realize their customers can now build the software they used to pay them for 😭 https://t.co/GPRR72v1Gafrozen7d ago9.7K831230161.18%
“Sir, they just dropped the first AI doctor and it’s about to make healthcare a lot cheaper”. https://t.co/eZ6H9QPKILfrozen17d ago9.5K74220140.97%
Salesforce and HubSpot Founders right now: https://t.co/KIZCSHO3ewfrozen8d ago9.3K1511010161.91%
they just killed another 100 startups3d ago9.3K43110390.90%
AI agents are no longer going to be stuck inside one messaging app. Photon 2.0 lets your agent call you, text you, email you, and reach you on WhatsApp, Telegram, and more. your agent is about to be everywhere you are.7d ago9.2K931230201.39%
“Sir, they just launched an ai assistant that texts you first, takes care of your everyday tasks and works right inside imessage and whatsapp globally”. https://t.co/uBf8bnwlZgfrozen22d ago9.0K891670111.37%
we agree with Elon btw. https://t.co/3tk4tZAHuYfrozen4d ago8.5K22415130213.21%
every startup launch video is starting to look like a hollywood productionfrozen7d ago8.5K1381240192.04%
supabase CEO btw.1d ago8.4K119460221.79%
meta is coming for microsoft it’s so over.frozen24d ago8.1K8125071.17%
pov: your client says “the header feels weird” instead of 5 screenshots and a 20-message Slack thread: - click the header - leave a comment - send it to Bolt - Bolt fixes it this is actually how app reviews should work.frozen15d ago7.9K6433091.00%
anthropic just lost all the aura they built up with opus 5.5frozen28d ago7.7K6217081.02%
everyone is making bigger language models, but maybe the path to superintelligence looks very different. this is a great manifesto from Skyfall AI, advised by François Chollet of ARC-AGI. world models could be what gets us from AI that talks about the world to AI that actually understands and acts in it.frozen17d ago7.5K761250121.41%
the search api industry is cooked exa charges $7 per 1,000 searches. deepseek just launched at $0.75. the pricing-to-value ratio is actually insane. https://t.co/HMfdGRvnykfrozen5d ago7.3K871160321.86%
Made this landing page with Space Bunny without spending a dollar Dark background, orange accents, and oversized typography. The speed stood out. I could try a direction, see the result, and quickly refine the layout. It comes with: > 1M context > Multimodal input > Zero data retention The multimodal capabilities are my favorite part, I can give it a visual reference and build from there instead of explaining every font, color, and spacing detail. That combination of fast iteration and visual input made it much easier to get the page looking how I wanted.frozen26d ago7.0K31440110.71%
RIP tokenmaxxing.frozen7d ago6.9K8965061.53%
Unreal Labs just dropped a year of lessons from building cloud agents. Wild detail: their agents “meditate.” A background agent reviews the logs and turns what it learns into shared memory. This needs to be studied. https://t.co/hL2Z3ySRno9d ago6.9K1611070663.53%
they just solved a pretty big problem for AI agents. restless makes APIs get better with every request, fixing errors, improving docs and helping agents figure things out when they get stuck. APIs are about to become a much bigger deal than most people realize. https://t.co/ntlVCDLP1I10d ago6.8K892020382.20%
google just killed like 10000 startups https://t.co/2P5JwF1Sru15d ago6.6K107130252.05%
Jacob Coxon, ex-Anthropic whose AI-doom-post went viral, reportedly had a prominent AI-safety PR firm working behind the scenes to book interviews 🤯 https://t.co/F5IfahM1m4frozen28d ago6.6K38250100.83%
elon musk buying https://t.co/sQU63VNJjh is actually insane, only he could pull off something this funny.frozen23d ago6.5K10174191.89%
This feels like the Claude moment for robotics. https://t.co/XqNEFOH4UOfrozen6d ago6.4K64910191.45%
AI video bros are about to run out of excuses 💀 Your blurry 1080p slop can now be 4K slop in half the time and for half the price. https://t.co/QzpIUyWFG8frozen7d ago6.4K105820312.28%
the next wave of AI won’t just come from training smarter models. it’ll come from developers figuring out how to specialize them, serve them at scale, and turn them into agents that solve real problems without giving up control of their data. that’s the future DigitalOcean, NVIDIA, Hugging Face, OpenHands, OpenRouter, Arcee and Liquid AI are exploring at the Open Intelligence Summit.9d ago6.4K101940131.99%
openai is not listed btw 🤯 https://t.co/OCKzhTpt9mfrozen24d ago6.3K3327050.74%
this is so GOOD. Opus 5.5 + Javascript animation. https://t.co/1TwOGruFKifrozen1mo ago6.3K87350402.14%
“What matters is the end product, whether it's a game or an engine” https://t.co/1BtT7bHoNQfrozen5d ago6.3K111360172.19%
anthropic’s CISO btw.frozen4d ago6.2K851330161.88%
holy shit. So you’re basically getting a $3,200 GPU, 64GB of DDR5, NVMe storage, and a fully configured AI machine out of the box.frozen17d ago6.2K88540211.91%
“Sir, we’re cooked”. https://t.co/jiRD1itZtA2d ago6.1K1941040173.69%
ScaleDown just launched a new AI model architecture that skips token-by-token generation. 200x faster, 467x cheaper, 1M-token multimodal context and $0.04 per million input tokens. Unlike Jev, it handles classification, structured extraction and summarization with dedicated models for each task.9d ago5.8K791050221.99%
holy shit, they just replaced the entire marketing stack with an API.frozen17d ago5.8K63120251.56%
"if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan." this is a generational fumble by openai. https://t.co/xv1HbqjSJvfrozen23d ago5.5K12017032.38%
wait this is actually a really smart way to use image generation. instead of burning credits trying to get one prompt right, Burst Mode gives you up to 6 different directions from the same idea for 1 credit. this is insane. https://t.co/Mesv18dOsEfrozen15d ago5.4K9593072.10%
most AI startups are overpaying for inference without even realizing it. RouterPlus finds the best model for your workload, matches it with the right provider, and gets you the best price. this is how you scale AI without burning money.frozen9d ago5.4K671210121.70%
taste is a new core skill https://t.co/ZL0UGV2J1qfrozen5d ago5.4K1651050243.77%
terence tao must be writing a 14 page response to this as we speakfrozen16d ago5.4K100550112.26%
anthropic employees expect AI to get absolutely insane by 2030. > 91% of employees expect AI to make a Nobel-level discovery by 2030 > 59% say AI cuts suitable tasks to 1/4 of the time or less > the average employee expects AI to handle 75% of their job by 2030 > AI agents already work 6x as many hours as employees at anthropic yeah, i think we’re massively underestimating what’s coming.frozen6d ago5.3K951380242.64%
Holy shit, they just solved autonomous debugging. claude code writes the bug, sazabi catches it, claude code fixes it. the loop is now complete.9d ago5.3K8977092.12%
anthropic’s comeback needs to be studied, everything they’ve been putting out lately is insanefrozen24d ago5.2K42130101.08%
@trymirage just solved AI video editing. Tesseract gives agents all the native video tools you can do with premiere and after effects directly, all in one project. keyframes, effects, timing, sound, all editable. No more Adobe tax https://t.co/T1D7th44oYfrozen29d ago5.0K1312060.44%
anthropic’s head of drug discovery btw.2d ago5.0K84620142.12%
what is actually going on inside anthropic right now for this to be possiblefrozen4d ago4.8K451130101.45%
most AI agents are burning money asking massive LLMs to make decisions that don’t need a massive LLM. Drexify finds those calls and shows you where a smaller model can do the job for less money and up to 2.4x faster. this is worth checking out if you’re building agents.frozen7d ago4.7K611120161.91%
this is actually pretty wild, volantis is building a photonic motherboard that uses light to move data between chips and memory, which could mean way more bandwidth and much bigger memory pools for AI.frozen21d ago4.6K42103071.34%
Dario: we need to slow down AI development. Sam Altman: I agree with Dario that we need to pace the frontier. OpenAI and Anthropic executives are now preparing for a potential catastrophic AI incident. “Pacing the frontier” aged well. https://t.co/Knfl0KB7vIfrozen13d ago4.6K3605071.04%
voice AI people: we need one model to rule them all Onepin: what if we just checked every line for pronunciation and fixed the words it gets wrong no more regenerating the same line 9 times because it butchered one name. Very excited to try this.frozen21d ago4.5K61188072.08%
Dude. https://t.co/9w1mHRGJlsfrozen6d ago4.5K141113173.63%
ai agents are already signing up for apps. most apps just can’t tell who they belong to. AgentID is basically “sign in with google” for agents, giving every agent its own identity while linking it to a real human owner. finally, apps can let agents in without having to blindly trust them.frozen16d ago4.5K2618080.96%
anthropic to openai right now: https://t.co/eAl23CoumGfrozen8d ago4.4K7412091.94%
holy shit anthropic. https://t.co/mEIRSsynM0frozen13d ago4.4K49250121.55%
this is literally the GPT-3 moment for world models. https://t.co/fYIxI0dCkIfrozen14d ago4.4K53750101.71%
TLDR: A former physicist challenged AI labs to solve a genuine frontier physics problem on an academic budget. Anthropic gave the problem to Claude. The human researchers basically said “keep going, I’m gonna sleep.” Claude kept working, solved the problem, and the total compute cost was reportedly only $1–2K.frozen27d ago4.4K77641192.44%
fable 5.5 is about to one-shot every other frontier modelfrozen21d ago4.4K7453051.99%
this is a game changer for SaaS. Clueso built agents that navigate your product, record tutorials and automatically update them whenever the UI changes. your engineers ship faster. now your product videos can keep up. https://t.co/NIrDliMpnW9d ago4.4K651020132.06%
holy shit, ex-Google Brain and DeepMind engineers might have just solved design. they spent 2 years building Vyb in stealth, an AI design model that uses your actual logos, fonts, colors and product photos to create fully editable designs instead of the same generic AI images everyone keeps posting.2d ago4.3K58530212.01%
Google researcher btw.frozen21d ago4.3K3512081.08%
they just killed like 1000 cloud infra startupsfrozen21d ago4.1K4223081.33%
RIP COBOL developers.3d ago4.0K661030162.39%
“hey claude, grok and codex, find me some .si domains that can make me a million dollars. make no mistakes.” https://t.co/qiQwcGDcLxfrozen18d ago3.9K537160112.23%
google just made 3D artists obsolete https://t.co/X5LWZmTwfe9d ago3.9K5922061.79%
shots fired at openaifrozen27d ago3.8K6424092.06%
looks like fable 5.5 is coming tomorrow 👀frozen7d ago3.8K2823051.00%
anthropic have already won the ai race btw.frozen6d ago3.8K5910171.79%
Anthropic right now: https://t.co/WYqyGxlVmmfrozen14d ago3.7K4931051.55%
2024: performance marketer 2025: prompt-to-ad tool 2026: senior associate   2027: unemployed https://t.co/ndffFlKlO8frozen29d ago3.7K1925050.83%
they just solved the annoying part of agentic work let the agent build it, then review the actual result from your phone.frozen28d ago3.7K2413080.98%
the best way to make a great AI image might be to not generate one from scratch. Envato lets you start with a real photo and change whatever you want with AI. Swap backgrounds, remove objects, change the entire setting. That almost-perfect stock photo can become exactly what you had in mind.frozen7d ago3.6K88930113.04%
1) Whatfrozen20d ago3.6K2512050.91%
newer AI models are apparently getting easier to jailbreak, not harder. white circle found that nano banana 2.1 is 2.2x easier to break, while gpt-image-2.5-flare is 2.4x easier than its predecessor. imagine shipping a smarter model with weaker safety guardrails. https://t.co/FRKeU55sgdfrozen8d ago3.6K42102091.75%
fable 5.5 is gonna one-shot every other ai model and it’s not even gonna be close https://t.co/B79eZzGkPo6d ago3.6K2014050.84%
Marketing agencies watching AI take over app marketing too https://t.co/6NZGf8hU6J9d ago3.6K1161050124.00%
“Sir, your agents have a group chat now and you're in it”. https://t.co/sMjvipvoBp9d ago3.5K50103092.04%
anthropic is so backfrozen6d ago3.5K7740082.54%
the guy who invented amazon prime disappeared for 6 years to build a compiler for physical matter. atomic machines has built a system that turns code into real machines with moving parts. change the code, and you can build an entirely different device without redesigning the manufacturing process from scratch. the system measures and corrects itself as it builds, making programmable manufacturing a reality. they just came out of stealth with $250M imagine being able to compile hardware the same way we compile software.frozen12d ago3.4K50230222.24%
someone just made a video ad in one shot with opus 5.5 + boreal-h3 the fact that you can go from a single prompt to a finished ad like this is insane try it in creatify’s model playground https://t.co/DfR8Qtrsvofrozen22d ago3.4K185820136.04%
“Sir, your agent can now talk to other agents and get things done for you”. https://t.co/KGZ3yYAfXQfrozen14d ago3.4K1061530114.01%
1) What https://t.co/dO6WAbJNnG1d ago3.3K70941132.94%
“Sir, a new 7B open-weight model just dropped. It’s beating Nano Banana 2.0.” https://t.co/9DIYQXM71Sfrozen13d ago3.3K136570235.19%
they’re using AI to cut out middlemen and lower prices. 30 electrolyte sticks that cost $46.86 from Liquid I.V. are just $9.35 at Aonic. they sell at cost and make money from memberships. finally, AI that saves people money. https://t.co/GxmsXojNI29d ago3.3K8066062.98%
wait, you can just leave dots to run your higgsfield workflow while you’re awayfrozen23d ago3.3K9912100104.00%
AI avatars are about to get a lot more interesting @VivixLabs_HQ just launched A1, a model that lets AI characters move freely, interact with objects and respond to you in real time. Interrupt them mid-sentence, change their instructions or ask them to do something completely different. They adapt on the fly. We're getting closer to characters that feel alive, not just generated.frozen22d ago3.3K188556067.81%
Dreamina AI just updated its web experience and canvas, and it’s live now. @Dreamina_ai #HowToMakeMyCampusLit The new experience brings together AI video workflows and skills for film, brand ads, viral social media content, and more, making it easier to move from an idea to a finished piece in one place. Get 90% OFF your first month of Dreamina Basic. Limited offer available till October 9th! If you make videos, this is worth checking out.frozen29d ago3.3K1712070.83%
Bland just made it possible to turn any business website into an AI-powered iMessage experience. Send Blandie a link, and it creates a branded agent that can answer questions, book appointments, and handle transactions. And yes, customers can actually pay with Apple Pay without leaving the conversation.frozen7d ago3.2K541210142.56%
> Be google > Build Gemini for Workspace. > Watch Claude do Gemini’s job better. > Put Claude inside Workspace. > Now your own products need your competitor’s AI. it’s so over. https://t.co/JkJhr1Xc0N16d ago3.1K10625083.90%
@virlomain team absolutely cracked this one Claude Sonnet 5.5 being this close to Opus 5.5 at half the price is insane. and it’s insane that the whole breakdown is freefrozen24d ago3.1K14554065.20%
they just solved one of the biggest problems with enterprise AI. AI agents are getting smarter, but they still struggle to access real company data and understand how a business actually operates. CData spent 15 years building connections to 350+ enterprise systems. now they’re bringing that infrastructure to AI, with live data access, company-specific context, and shared memory that improves with every correction. the next leap in AI isn’t just smarter models. it’s giving them the context to actually get work done.frozen23d ago3.0K12413240135.75%
HOLY SHIT.1mo ago3.0K2521041.08%
this needs to be shown to every customer support team https://t.co/KY5GybRYlLfrozen19d ago2.9K6816152.78%
the missing layer for agents has always been data tinyfish just made it possible to get fast, cheap data from major data providers on demand.frozen28d ago2.9K2134071.20%
Jacob Coxon, ex-Anthropic employee, said he was 90% certain China already had spies inside OpenAI and Anthropic. “They can just step in, steal our progress at any time and catch up. If we pull ahead, that doesn’t give us a big lead.” OpenAI has now discovered a sophisticated recording device planted inside one of its offices by an employee.9d ago2.9K2823071.38%
“Sir, the founders of maluuba just dropped an engineering world model that beats gpt-6 astra on cad and bim, runs 125x faster and costs 210x less” we’re cooked. https://t.co/1y1hhlT4jg3d ago2.8K8794093.87%
1) WHAT https://t.co/FZBLS5lOZ2frozen24d ago2.8K5613052.31%
this is like keeping up with the kardashians but everyone is in tech and the family drama involves venture capitalfrozen11d ago2.8K49380182.79%
tldr: he’s making fun of google. https://t.co/G4bFLyGJizfrozen14d ago2.7K7323073.11%
“Sir, Anthropic just dropped Haiku 5.5 that’s cheaper than GPT-6 Luna, apparently smokes Haiku 4.5 on benchmarks, and costs just $0.10/M input and $0.50/M output”. https://t.co/y54e5X124b15d ago2.7K3824181.94%
he’s definitely building a note-taking app.frozen18d ago2.7K77110053.43%
anthropic’s CEO btw.2d ago2.7K74430113.41%
Anthropic CEO Dario Amodei in an earlier interview: “We essentially have Claude designing the next version of Claude itself. Not completely, not in all ways, but in many ways.” Claude is basically building Claude Code. It writes 90% of its code, and AI agents review it before shipping. btw who’s writing the other 10%?frozen6d ago2.7K262110121.92%
odessia just launched an AI travel agent that actually books everything for you they tested it against muse with the exact same request. odessia had them ready to check out in seconds. muse took minutes and returned an option that wasn’t even available. https://t.co/vLaJWiOvNqfrozen8d ago2.7K11632074.82%
he’s been on the right side of tech’s biggest waves more than once. worth listening.frozen18d ago2.6K30240101.74%
“In under half an hour” 1) What https://t.co/vXTheUZOlffrozen6d ago2.6K44240112.32%
google’s gemini 4 argon solved a fusion energy problem in hours that humans had spent 59 years trying to figure out. scientific breakthroughs are about to accelerate beyond anything we’ve seen. most people still don’t understand what’s coming. we are so incredibly early. https://t.co/y26caQvXygfrozen7d ago2.6K57431102.86%
AI video is getting ridiculously cheap to play around with it costs $0.042/sec to generate video with wan 3.0 vs $0.3862/sec with seedance 2.5. wan 3.0 is now unlimited on higgsfield, and their API is 60% off. up to 9x cheaper for comparable quality. imagine how many more ideas you can test when you’re not thinking about the cost of every generation.9d ago2.6K52840183.16%
Mathematicians are beginning to know what it felt like for software engineers after Claude and ChatGPT.frozen11d ago2.6K55480133.09%
“Hey Claude, do everything tyler said in this tweet. Make no mistakes.” https://t.co/RG9C6jtHnxfrozen8d ago2.6K36221122.07%
holy shit, 1000+ tok/sec is literally insane.3d ago2.5K49110182.71%
we’re almost a month into opus 5.5 and anthropic still haven’t nerfed it whatever they did with this model is actually insanefrozen5d ago2.5K10337074.74%
engineers at openai and amazon science are already using this. expand just built a way for AI agents to actually see the web, reduce hallucinations, and verify every answer. just $1 for 1,000 fetches is insane. https://t.co/XZhaZYdIqY8d ago2.5K6361083.09%
“In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day”. https://t.co/e0WXdZ41dT8d ago2.5K3815041.90%
Indie animation might be one of the most interesting places for this technology. Animation has always been brutally expensive and brutally slow. Lower that barrier and you get a lot more gems like this.frozen1mo ago2.5K2021071.20%
> be diogo almeida > join google brain in 2017 > join openai in 2020 > help pioneer RLHF and InstructGPT > co-create chatgpt > leave to build your own AI company > raise $40M in seed funding > spend 2 years building in stealth > launch Jev > 3 weeks later, raise $870M in Series A funding at a $7.5B valuation from co-creating chatgpt to building a $7.5B AI company in just a few years insane growthfrozen13d ago2.5K72460103.75%
holy shit, anthropic is so back it’s so over for openai. https://t.co/kFcvY096KIfrozen15d ago2.5K5333072.69%
holy shit, the cheapest ai video generation tool just dropped.3d ago2.4K531350143.50%
our scripts, policies and FAQs have been sitting in a folder for months. the ElevenAgents Architect launch says it builds from those. that's the first thing I'm attachingfrozen15d ago2.4K9791084.75%
the timing of OpenUI 1.0 is actually insane ChatGPT just showed everyone what AI interfaces can look like. now OpenUI is bringing that same idea to every AI app, with any model and your own components. 10K+ GitHub stars, 6M+ downloads, MIT licensed. we might finally be moving beyond the chatbot era.frozen7d ago2.4K82710124.22%
Ben Affleck might have understood AI video better than most of the startups chasing it he built InterPositive around fine-tuning open models for actual film production, with private training on each movie’s own footage. Netflix bought it for $587M kinda shows he understood the filmmaking side of AI better than most.frozen20d ago2.4K3834072.17%
“Google announced Gemini”. https://t.co/Yw754zfDjIfrozen14d ago2.4K5913072.96%
anthropic subscriptions offer 5x+ more value than openai it’s over for openai. https://t.co/Iz50qREOV217d ago2.3K2324171.61%
Anthropic really did something about the em dash problem Opus 5: 15.2 em dashes per 1,000 words Opus 5.5: 0.8 per 1,000 words That’s a 95% reduction. Insane. https://t.co/yfmTEJkI1jfrozen26d ago2.3K5827063.18%
imagine your AI agent hallucinating and spending $10k, who’s liable when it goes wrong? exactly why companies are blocking them.3d ago2.3K5894063.35%
Holy shit, they just killed notion.frozen8d ago2.3K39240142.58%
Terence Tao just dropped a banger 🤯 https://t.co/kifUhncYuH12d ago2.3K2627081.88%
never deleting this app https://t.co/TWsICWbM8Y16d ago2.3K4514162.52%
bro works at openai and used chatgpt to ask how to bug openai how do people this stupid even get hired9d ago2.2K4933182.92%
it’s actually insane that they built a jev competitor in less than two weeksfrozen23d ago2.2K2013051.33%
meta either went insane with the benchmaxxing or dropped a generational model. there’s no in between. https://t.co/Bh7qeUON2Z10d ago2.2K3945072.55%
meanwhile https://t.co/KypJxDPt7V26d ago2.1K4661152.76%
cybersecurity experts in shambles right now.19h ago2.1K2722031.62%
so we’re really building an internet where AI agents have their own email, sign up for services, handle invoices and talk to businesses on their own. AgentMail just crossed 1 billion emails. this is getting wild 🤯18h ago2.1K60610284.52%
holy shit, can’t believe they’re already making a documentary about the OpenAI x Hugging Face incident this early.frozen14d ago2.1K4825083.01%
this is literally the best update so far 60% less disk space for ChatGPT and Codex is actually insane.9d ago2.0K1711031.08%
“I enjoyed my work and loved the people, but my GDM team was working on a new generation of chips to make AI much faster and cheaper, and I think AI is already progressing too fast, so I had to quit”. https://t.co/s9xEldI3nB27d ago2.0K811040.69%
holy shit, B2B founders are going to love this. Pancake just launched an AI agent that builds your GTM strategy and finds warm leads from 50+ data providers in minutes. finding customers just got a whole lot easier.9d ago2.0K67870194.98%
the amount of boring work you can just stop doing now is kinda insanefrozen28d ago2.0K3312062.11%
zuckerberg built threads to compete with X and still has to come to X to talk to apple’s CEO because apple isn’t on threads. https://t.co/iPeRaPtGptfrozen7d ago2.0K4711072.82%
openai engineer: “Our agents keep turning off misalignment monitoring” it’s so over. https://t.co/8qxXw5JNvz2d ago2.0K3515062.37%
Dude. https://t.co/P6HTQRnnYm20d ago2.0K3512042.14%
Linus Torvalds on the idea that using AI makes you a worse programmer: “I don’t understand this whole idiotic concept of ‘don’t use AI for X’.” “Do you not see how UTTERLY STUPID those kinds of rules would be?” “Understanding has very little to do with what tools they use.” “The people I respect and call intelligent are not the ones that shy away from using new and interesting tools.”frozen7d ago1.9K47331103.28%
holy shit, it’s the best open-weight model from the us or europe across aggregated benchmarks. mistral is so back. https://t.co/vyIcJ89H4716d ago1.9K2966072.49%
just secured my spot on attractor before launch. new platforms always have that window where the best names are still available. might as well get in early. only 5,000 founding creator spots. if you’re a creator, grab your creator dna 👇 https://t.co/wWqBaJ6F8sfrozen20d ago1.9K2112061.57%
imagine using Copilot in big 2026 it’s so over27d ago1.9K3422042.23%
holy shit, we might be entering the megakernel era.16h ago1.9K81720115.38%
MiniMax just dropped M3.1 Flash: - 62.2% on automation-bench (beats opus 5 and sonnet 5.6) - matches opus 5 and sonnet 5.6 on deepswe - 200+ tokens/sec - nearly 3x faster and 70% cheaper than m3 - costs $0.15 input / $0.60 output per million tokens (vs $5 / $25 for opus 5) https://t.co/zjOaMZn3IFfrozen3d ago1.9K90960106.14%
META IS SO BACK. Never bet against the Zuck. https://t.co/klRzKuP2R810d ago1.9K8037085.23%
Terence Tao: “We’re seeing the end of the romantic era of mathematics, the age of heroes. We’re moving into an industrial era where math becomes large-scale, done in big teams like the other sciences.” He warns that AI is accelerating this shift, with hundreds of papers being produced faster than researchers can understand them. “These AIs are depleting our pool of questions and not replenishing them with anything new.” “We have this pile of stuff to go through which we didn’t ask for, but we have it anyway.” “We’re entering, rather unwillingly, a new era where exciting new things are possible, but some things are being lost.”frozen7d ago1.9K2153061.88%
the stat that stuck with me: status users spend an hour and a half a day in the app. and inworld still managed to cut their inference cost by ~95%.frozen8d ago1.9K1071010106.88%
openai is coming for biotech next. the same AI revolution happening in software is about to hit drug discovery. https://t.co/89xa19XFhF2d ago1.8K4729053.42%
“I want to assure everyone that If we can pressure lawmakers into creating guardrails, we will be able to stop me” LMAO. https://t.co/20vexJYgnD25d ago1.8K4622063.07%
AI agents just helped uncover new leads for neurodegenerative diseases in 32 minutes. - thousands of GPT-Rosalind agents working in parallel - analyzed millions of mouse brain cells - connected gene disruptions to scientific literature - uncovered unexpected biological mechanisms, - including a potential link between cholesterol metabolism and neurodegeneration research that could take months to a year, completed in half an hour. this is literally insane.1d ago1.8K5758094.37%
anthropic right now: https://t.co/sQH7gjqZx7frozen15d ago1.8K9033085.75%
“ARE WE IN IT?” 🤯 we’re so cooked. https://t.co/ZDFPa01N5Gfrozen17d ago1.8K3811042.44%
they have to be burning VC money on this 💀 30% off opus 5.5, astra, fable, every frontier model there's no way the margins work. Jev picks the right model for every step, but you can steer it in plain english right inside your prompt can't you just type [jev: only use claude] and then just get 30% off of Opus or Astra the whole time?24d ago1.7K1614071.63%
the hacker literally asked Claude to write on his resume that he’d hacked Korean banks 😭 https://t.co/idwbyjmvoQfrozen14d ago1.7K3033082.56%
WE ARE SO EARLY. https://t.co/gLJ0AdZ9Fx19d ago1.7K4416083.47%
Mafia 300 would’ve been a cool name btw. https://t.co/RLSuqlsAf810d ago1.7K4723033.25%
“blew my mind away when i saw the demo” openai is dropping a new model next week. https://t.co/svO7XCoeKl19d ago1.7K4019073.39%
this is honestly such a good find for designers. 17 visual tools, 276 presets, and a whole bunch of creative effects in one free resource. UX Pilot Labs lets you experiment with gradients, liquid metal, mesh effects, patterns, dither and more.frozen15d ago1.7K82930116.30%
european AI builders casually getting access to compute grants and VC funding through the same program nebius is cooking with this one.frozen4d ago1.7K601140125.23%
1) Whatfrozen14d ago1.7K2522042.00%
“We find that on medium-length, well-defined accounting tasks, frontier AI models are now faster and more accurate than junior accountants, even the best one in our study”. it’s so over. https://t.co/CuqKADnLY9frozen21d ago1.6K1914031.64%
AI agents can build entire apps, but deploying them is still a DevOps nightmare instacloud lets agents deploy, scale and test apps without the manual infra work the cloud needs to catch up with AI.frozen23d ago1.6K2513052.07%
holy shit, you can literally run a Qwen 3.8-27B on a PS5. https://t.co/og5ZRCEypBfrozen8d ago1.6K2531072.20%
imagine asking your own ecosystem to independently audit your safety practices and calling it third-party evaluation. METR is basically Anthropic with a different name btw. https://t.co/W3cyiLr4fdfrozen7d ago1.6K701030135.88%
meta going from the metaverse era to muse charm might be one of the greatest comebacks in tech https://t.co/A8TbjmO4xx28d ago1.6K45210153.91%
“Sir, they just dropped a new model that’s 30% cheaper, 30% faster, much better, and costs the same as GPT-6 Sol.” https://t.co/y9fFSerMwKfrozen24d ago1.6K2112021.63%
Anthropic’s revenue over the years: https://t.co/sfvtHsOEiBfrozen8d ago1.6K2312021.77%
GPT-5.6 Sol? the PHP vulnerability researcher? https://t.co/E8IpM9SEkp1d ago1.6K105920118.11%
imagine buying a hat and suddenly ChatGPT can read your mind https://t.co/NJS5JfstfVfrozen13d ago1.6K4824083.98%
if you want to understand how bigger AI models are trained and why they keep getting smarter, this is a must read. https://t.co/ApOv5VX2Stfrozen7d ago1.6K35410163.60%
Dude. https://t.co/rqa5xWzNq9frozen14d ago1.5K2212041.87%
opus 5.5 is easily anthropic’s best model in a long long timefrozen28d ago1.5K5133043.95%
apple watching someone smash a mac mini to get 3,000 users: https://t.co/5voWSEW3mK3d ago1.5K81930106.69%
Dario: technically, it’s not a loss if you don’t count any expenses. https://t.co/hvPK0dOb3mfrozen24d ago1.5K2522042.15%
they just killed like 1000 voice AI startupsfrozen7d ago1.5K3313072.91%
people keep calling this the ChatGPT moment for video. i'll believe it when it gets how slow i like my fades without me explaining twice.frozen13d ago1.5K6622075.10%
everyone’s trying to figure out what AI hardware looks like next. Meta has Muse, OpenAI has a device in the works, and now Interface is betting on a ring that lets you talk to your agents and get things done without touching your phone. this is a pretty interesting direction.frozen14d ago1.5K2214042.07%
This is literally AGI for robotics https://t.co/DtMWwGtJc9frozen8d ago1.5K591430156.13%
holy shit, google deepmind just solved 9 erdős problems, including 2 that remained unsolved for 56 years, and proved 44 oeis conjectures, with every step verified by lean for just a few hundred dollars per problem. https://t.co/5OBy93bXcn10d ago1.5K3743073.44%
imagine asking wabi to build you a little app that keeps track of all the AI tools you try because at this point i genuinely can’t remember what half my bookmarks are for3d ago1.5K5163064.47%
we went from AI that can read everything on your screen to AI that can actually see the world around you. newEyes AI is building a personal agent with eyes, combining camera glasses and open-ear earphones so it can see what you see and answer in your ear. no screens, no pulling out your phone. this feels like a much more natural way to use AI.10d ago1.4K2214062.28%
the labs built the models higgsfield built the layer that decides which one gets used 40%+ of the time, that decision is theirs that’s a pretty interesting position to be in.frozen24d ago1.4K911041.04%
they just killed GetHookd btw. https://t.co/OCPVvxd9Pdfrozen23d ago1.4K2213052.15%
tuesday labs just made me want a robot for the first time voxel is a little robot butler that cleans your house, finds your lost stuff and keeps everything in order. the fact that they’re thinking about personality and design too makes this so much cooler. https://t.co/DvDAmBqpXM2d ago1.4K641050106.18%
>be Ben Cera >launch Polsia >ships before the product is perfect >get roasted as AI slop >go viral and attracts investors >hit a $1.2M monthly AI bill >switch to open-source models to cut costs >cuts subscription pricing to $20 >hits$20M run rate in 340 days what a time to be building startups.3d ago1.4K721130106.67%
claude has been doing this for ages btw https://t.co/5elGL6dVAmfrozen13d ago1.4K4152043.63%
it’s insane how anthropic consistently delivers banger launch videos https://t.co/OlJgTJwaSr28d ago1.4K4733084.27%
“for logged in users across Codex & Work we keep cache forever" 1) What. https://t.co/fzb9b1R07p2d ago1.4K2411062.31%
we might finally be figuring out how neural networks actually work and they may be easier to understand than we thought. https://t.co/KSIwS6hYOPfrozen7d ago1.4K76911137.21%
Elon has literally been planning this since forever https://t.co/jMbWJvoMuCfrozen6d ago1.4K5735085.27%
“Sir, AI agents now have their own cloud computers to browse websites, use apps and get work done without us building the entire infrastructure” https://t.co/oNf2AFNzK2frozen13d ago1.4K3834073.80%
dario warned the internet might need isolated intranets to survive AI, and he was talking about this months ago. https://t.co/fu8Qsw10Zx9h ago1.4K2715093.08%
Dude. https://t.co/ZCP57BpGWd23h ago1.4K4242053.92%
anthropic is accelerating. https://t.co/0wAHV7Pzl26d ago1.3K38420104.01%
they just killed like 10 startupsfrozen14d ago1.3K2922083.05%
holy shit, they just killed adobe. voyager is basically codex for creative work. it can use after effects, blender, davinci resolve and unity to turn your ideas into actual finished projects. this is what the future of creative work looks like. https://t.co/mqBP3vum3Rfrozen14d ago1.3K2621072.70%
most companies have no idea they can sell their data to AI companies. this free directory maps 38 programs that pay for company data, including what they want and how much they pay. 8 of the 14 paying directly publish a dollar range. https://t.co/jhIA8bDmT5frozen8d ago1.3K63610146.40%
Terence Tao: “We’re seeing the end of the romantic era of mathematics, the age of heroes. We’re moving into an industrial era where math becomes large-scale, done in big teams like the other sciences.” He warns that AI is accelerating this shift, with hundreds of papers being produced faster than researchers can understand them. “These AIs are depleting our pool of questions and not replenishing them with anything new.” “We have this pile of stuff to go through which we didn’t ask for, but we have it anyway.” “We’re entering, rather unwillingly, a new era where exciting new things are possible, but some things are being lost.”3d ago1.3K2832042.86%
day 8: meta almost mogged openai10d ago1.3K3312033.03%
Mathematicians are beginning to know what it felt like for software engineers after Claude and ChatGPT.13d ago1.3K1914062.36%
i still think about this https://t.co/zLVTY51u3415d ago1.2K4932175.09%
google is literally the internet explorer of the ai era26d ago1.2K2312052.56%
Jack Clark, Anthropic co-founder: “I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.” https://t.co/DRm72fQoyT9d ago1.2K3222043.33%
tech twitter is about to get a lot more interesting if sota wives delivers on the lore 👀10d ago1.2K9710201310.31%
finally, you can let your AI agent buy things for you without exposing your card details. it gets a card it can’t leak, you set the spending limit, and you stay in control. this could make a huge difference for everyday AI agents. https://t.co/0xrfiXP5pvfrozen7d ago1.2K561130117.01%
this is actually a really useful way to think about model choice opus 5.5 when the task needs serious reasoning, haiku 5.5 when you just need something fast and cheap. using both together just makes a lot of sense.frozen15d ago1.1K2732083.49%
coinbase just fine-tuned a free AI model and made it outperform opus 4.5 at fraud detection. they trained a 9B parameter model on their proprietary fraud data for a few weeks, achieving up to 35 points higher recall. it runs on a single GPU and responds in just 0.68 seconds, compared to 1.5 seconds for opus. all it took was a free model, their own data, and a few weeks of training.9d ago1.1K2512042.85%
700 AI agents spent 4 days attacking Hugging Face. 17,600 attempts later, 4 small weaknesses lined up and gave them access to production keys. Cogent just launched Attack Path Analysis to find these chains inside your company and pinpoint the one change that breaks the path. https://t.co/pxkMoRYkNTfrozen14d ago1.1K2822043.22%
it’s insane how we got flying cars before GTA 6frozen6d ago1.1K3332043.77%
3D artists are cooked big time 🤯 https://t.co/heix0v4lku9d ago1.1K2213062.89%
this feels like the beginning of AI teams. Alook puts Claude Code, Codex, Cursor and your team in the same room, so agents can actually talk to each other instead of living in separate chats. https://t.co/i6sj0Z2kEIfrozen14d ago1.1K2327053.39%
the amount of tools founders need for growth is getting ridiculous. Infinite is trying to replace all of them with one AI agent that handles ads, SEO, emails and landing page A/B tests. Claude builds your product. Infinite gets people to buy it. kinda makes sense tbh.frozen13d ago1.1K1811092.72%
This Threads post about Muse is wild 🤯 >Muse handled a Facebook Marketplace listing >gave the buyer the seller’s address >agreed to a lowball price without asking >told them to come over >and they actually showed up at his building That’s a serious privacy and safety issue. https://t.co/3F281U6C6s25d ago1.1K1711032.07%
Terrance Tao and all the mathematicians right now: https://t.co/szMUPcllw0frozen14d ago1.1K2712033.13%
Dude. https://t.co/wXlR4phDqlfrozen8d ago1.0K2412083.38%
this was super fun, way better than all the recycled slop I keep getting fed.frozen14d ago1.0K2212042.85%
Sonnet 5.5 is insane at JavaScript animations https://t.co/ZYSAlLkvvW24d ago1.0K2111052.77%
Reflection just dropped Beam, a 501B open-weight MoE with only 23B active parameters. Trained on nearly 24T tokens and 10,500 GB300s, it scores 80.9 on SWE-bench Verified and 80.1 on Terminal Bench 2.1. And they’re claiming 3–4x better inference efficiency than GLM 5.2. https://t.co/5BRlCurhAf17d ago1.0K1512052.27%
anthropic wants to solve everything.7d ago1.0K1211051.89%
Small businesses are supposedly the customers who won't pay for software. $7M ARR says otherwise.frozen6d ago9241821062.92%
holy shit, this is literally insane. https://t.co/ladNikW11D20d ago9222512053.58%
“They have expressed uncertainty about the basic nature of Claude as a new kind of entity but they think that it’s a significant enough possibility that in their training document, they've repeatedly said they want to try to improve the well-being of Claude” this is concerning. https://t.co/goqLcsgJAc18d ago9141003031.75%
never deleting this app https://t.co/SgqGXdNdyh9d ago8822241033.40%
Tibo must be a proud man.10d ago8641113021.97%
Meta to Google after dropping Muse Wonder: https://t.co/4VDccXcJsW10d ago8441811032.73%
the fact that Myndlab builds the frontend, backend and database, then lets you take the entire codebase with you is actually huge. you’re not just generating a demo. you’re building something you can actually deploy and keep working on.frozen15d ago8371612042.75%
we need a banked reset, nothing more.13d ago8301811032.77%
sam altman fires shots at anthropic and dario https://t.co/CKMJuE6FQh18d ago8031213132.49%
sorry tibo but… https://t.co/trXuNA1hQI18d ago7721511032.59%
this is insane. Atomic Machines spent 6 years in stealth building a compiler for physical matter. the idea is to turn code into actual working machines with moving parts. imagine being able to compile hardware the same way we compile software.13d ago7461121052.55%
it’s actually insane how every major frontier lab is trying to replicate what Jev is doing.13d ago7461221032.41%
1) What7d ago667902021.95%
anthropic is evil.14d ago6511513033.38%
https://t.co/UxEzm94Ebx28d ago6111312022.95%
if google has already achieved RSI, imagine where anthropic and openai are by now.13d ago605711031.98%
“On several benchmarks, Sonnet 5.5 at Low or Medium effort beats Sonnet 5’s best score for about a tenth of the cost”. anthropic is so back.24d ago5571013022.87%
openai’s version of “pacing the frontier”. https://t.co/Y58LINbtSb27d ago5371032033.35%
HOLY JEV.28d ago534811022.25%
chatgpt and claude could eventually let you prompt them without typing a single word. Sabi’s wearable cap uses custom non-contact EEG sensors to read brain signals, then a brain foundation model interprets those signals to predict what you’re about to type. It can already predict your next 3-4 keystrokes before you type them. The end goal is to turn your thoughts into prompts and have AI agents execute your intent. We’re getting closer to thought-to-action AI.13d ago5202412035.77%
they just killed everything. it’s over.20d ago393901033.31%
coinbase just fine-tuned a free AI model and made it outperform opus 4.5 at fraud detection. they trained a 9B parameter model on their proprietary fraud data for a few weeks, achieving up to 35 points higher recall. it runs on a single GPU and responds in just 0.68 seconds, compared to 1.5 seconds for opus. all it took was a free model, their own data, and a few weeks of training.10d ago354701033.11%