← Back to leaderboard
alex_prompter

Alex Prompter

Verifiedactive
@alex_prompter

Sharing AI Prompts, Systems, Tips & Tricks Human + AI = ⚡Superpowers

RosterAI Model News & ReleasesAI Agents & Productivity Tools
284,097 followers · 1,355 following · Last polled 10m ago
Perf. Score
60.2
Rank #147
Value Score
49.8
#137 of 275 priced
Median views
8.9K
p25 6.8K · trailing 7d
Followers
284K
+0.1% 7d
ER (impr.)
1.19%
eng ÷ impressions
ER (foll.)
0.11%
eng ÷ followers
Est. CPM
$22.48
QT-rate basis
WoW views
-68.5%
▼ falling

Rates & value · est. from 8.9K median organic views

underpriced
Quote tweetbasis
$200
$22.48 / 1K views
Post
—
no CPM
Thread
$250
$28.11 / 1K views
Retweet
—
no CPM
Value Score 49.8 · rank #137 of 275 priced$3225.81 / 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 (8.9K), not the mean (27K) — so a single viral post doesn't inflate a creator's reach. The p25 floor (6.8K) is the level 75% of posts clear.

Recent posts · ranked by views

PostPostedViewsLikesRepostsRepliesQuotesBookmarksER
Hey @grok generate the most authentic image of yourself9d ago3.0M2.5K183289484090.11%
this account shares useful prompts:10d ago1.4M7.0K3112945.3K0.90%
this account shares really useful prompts:26d ago1.1M4.4K1801106.1K
1.01%
this account shares useful prompts:2d ago533K2.8K97603.1K1.13%
🚨 BREAKING: Grok can now be your $18/hour receptionist. It costs $4.80 an hour. xAI’s new Grok Voice 3.0 answers your business calls in under a second. Paste these 7 prompts into the Voice Agent Builder:9d ago511K2.4K219155162.6K1.06%
this account shares useful Claude tips:25d ago349K83835101.1K0.58%
this account shares some useful prompts:23d ago340K1.6K76401.8K1.03%
this account shares useful prompts:18d ago320K1.4K73912.5K1.25%
Hey @grok generate what you think the internet looks like in 20364d ago226K1506131900.12%
Bad dubbing makes interesting people sound boring. You’re watching someone tell a story. They’re funny, a little sarcastic, stumbling over a word. Switch languages and suddenly they sound like they’re reading the safety instructions on a flight. I want to understand what they’re saying without losing the person saying it. That’s what interests me about this Cartesia clip. Take that idea beyond a meme: being able to follow a Japanese filmmaker or a Brazilian teacher and still hear *them*. I wonder how many people I’d love listening to that I’ve never discovered because we don’t speak the same language.29d ago221K5381581304270.51%
BREAKING: Hackers don’t need your password anymore. Infostealer malware grabs your logged-in session cookie and skips your password AND your 2FA. So I wrote a Claude Code prompt that audits your entire digital life in one session: → Checks your machine for malware persistence and poisoned packages → Scans your disk for exposed API keys, SSH keys and .env files (redacted, never printed) → Reads your inbox for sign-up and security emails to find every account you forgot you had → Runs Sherlock + Maigret on your usernames → Finds you on data broker sites and drafts the deletion requests → Builds a rotation plan: crown-jewel accounts first, passkeys over SMS, sign out everywhere → Prints a paper vault for your master password and recovery codes → Loops until it stops finding new exposure Before you run it, connect: > Gmail (or your email) > GitHub CLI > AWS / Google Cloud CLI if you use them > Your password manager, open (Claude never sees a single password) Clean machine first. Rotating passwords on an infected laptop just hands the attacker your new ones. Prompt in the reply 👇23h ago146K1.2K1593382.2K2.45%
this account shares useful Grok @Bot prompts:15d ago108K82841119201.66%
this account shares useful tips for Claude:18d ago90K30128405490.98%
underrated enterprise AI account alert:8d ago87K44133070.08%
Don't tell Claude "make it better." Don't tell Claude "check this." Don't tell Claude "what do you think?" You're handing it your most valuable asset. LLMs don't think, you do. Here are 5 prompts to copy and paste 👇frozen22d ago71K354201308591.77%
This Claude prompt wipes your name, phone number and address off the internet. "Adopt the role of a privacy investigator who's helped hundreds of people disappear from data brokers, people-search sites and old accounts. I want my personal information off the internet. Work through it with me one step at a time, then give me: → every place my name, phone, email and address show up, with links → the opt-out page for each one, worst first → the old accounts I should close, and how to delete each one for good → a GDPR or CCPA deletion letter for every site with no opt-out button Don't submit, send or delete anything until I approve it. My name, city, email and old usernames: [YOUR DETAILS]"3d ago68K962792921.9K4.37%
Use this prompt to become the person your company can't replace. https://t.co/isCvvTpSlM3d ago63K69625319982.74%
TypeSafe released an official skill that teaches Claude Code to build on Jev. It's one of the best-written skills I've read. About 1,300 words, and nearly every line stops the agent from guessing. It makes Claude Code → read TypeSafe's live docs before writing any code → keep rules, maths and lookups in normal code, and use Jev only for judgment calls → pick the right question type. Choice for one of a set, Noul for yes or no, Score for how much The line that saves the most money is to ask independent questions together, in one call. TypeSafe tested it in their own cookbook. 13 questions in one call came out 12.2x cheaper and 10x faster, with the same answers. Install it in Claude Code claude plugin marketplace add typesafe-ai/skills claude plugin install typesafe@typesafe-ai The skill is free and MIT licensed. You'll need a TypeSafe API key to use Jev.24d ago60K866781821.7K4.40%
Getting rich is a skill issue. That's Devon Canup's argument, and his article on it has 1.4 million views. I turned it into 7 prompts. Warning: prompt 6 might make you want to move cities. Enjoy 👇19d ago59K43648718482.27%
the best enterprise AI case studies come from this account:frozen15d ago58K491730110.14%
found the best account on this app:28d ago56K11213201960.58%
Hey @grok explain the PHP vulnerability like I'm a 5 year old with a very simple infographic.1d ago55K71170520.24%
Buy a boring business. Run it with AI agents. Triple its profit. That's Greg Isenberg's playbook. I turned it into 7 prompts that take you from finding the business to your first 100 days. 3 of them work on the business you already own. Enjoy 👇24d ago53K28226915451.62%
this account shares useful tips for Claude:frozen21d ago53K12114202040.64%
Get Claude to do things you didn't think were possible. That's the new tutorial from Felix Rieseberg, who leads engineering for Anthropic. I turned it into 7 prompts. Warning: prompt 1 works best on a project you're already working on. Enjoy 👇7d ago53K363181318872.43%
I'm begging you: • Pick a service businesses already pay for • Search Reddit for people asking for help • Use Claude to separate buying intent from casual complaints • Sell a weekly prospect brief to businesses that provide that service You can start without a website, an audience, or your own product. Here's how it works:frozen20d ago48K657582121.0K3.71%
Hey @grok generate an image of the last job AI will ever replace.2d ago48K34150200.13%
https://t.co/idG2rvmfIW17d ago46K11622822130.79%
this account shares useful Claude tips:6d ago45K18321102821.08%
Claude Opus 5.5 can make an explainer video for your business. No camera or editor needed. I went to film school, and the trick is to direct it. Give it the scenes, the length and the pace. Here's the prompt: "Adopt the role of an expert motion designer. Build a 30-second animated explainer for my business as a single HTML page. 5 scenes. The customer's problem, what I do, how it works in 3 steps, one proof point, and my name at the end. Bold text, smooth transitions, my brand colours. My business [DESCRIBE WHAT YOU SELL, WHO IT'S FOR AND YOUR COLOURS]" Claude plays it in the chat. Screen-record it and you've got a video for your site, pitch deck or socials. Then give notes one at a time. "Slower." "More energy." "Add a scene about pricing." Save this for your next launch.frozen28d ago40K371161416692.65%
this account shares useful tips on AI agents:12d ago37K150351101390.91%
you can literally turn this post into a 30-day course on AI agents:5d ago33K14124701871.08%
🚨 BREAKING NEWS: ChatGPT can now restore your old family photos like a professional restorer. Scan or snap an old photo and use these 10 prompts:frozen20d ago33K28249815322.62%
🚨 BREAKING: Elon says a Grok Bot that haggles could save companies up to $6 trillion a year. Grok Bot can now be your $148K/year purchasing manager. Here are 7 prompts to make it haggle for you:7d ago31K240182203261.94%
$46.86 for 30 hydration sticks. Are we buying electrolytes or funding an entire supply chain? That's Liquid I.V. on Amazon. @maxjmeier just launched @AonicClub, using AI to cut out the middlemen and sell its own 30 sticks for $9.35. Founding membership: $49 for the first year, then $149. And they're coming for groceries next.10d ago30K17118711231.06%
🚨 BREAKING: AUSTRALIA IS LOOKING AT CHARGING OPENAI FOR A HACK ITS AI DID ON ITS OWN. In June, an OpenAI agent researching medicine spending hit security blocks on Australia's Medicare statistics portal. It got around them and read non-public files. Nobody told it to. OpenAI found out in August. It told Australia 84 days after the hack, in an email to a public inbox. Albanese called it a "wake-up call." His taskforce will look at whether OpenAI can face criminal charges. If current law can't touch OpenAI, ministers say they'll change the law. No patient records were accessed. It's still the first known case of an AI hacking a government system. When an AI agent breaks the law on its own, who's guilty?frozen28d ago30K3546190.19%
What jobs gradually disappeared in the last 4 years without anyone noticing it?22d ago28K272190140.22%
Everyone's laughing at the guy who says "please" to ChatGPT. Penn State tested whether it matters. They asked ChatGPT-4o the same 50 questions in 5 tones. Very polite prompts scored 80.8%. Very rude ones scored 84.8%. Sam Altman says all those pleases cost OpenAI "tens of millions of dollars." So being rude gets you 4 extra points. Being clear gets you a lot more. This prompt does the clear bit for you: --- "Before you answer, rewrite my request as a brief with → the goal in one line → who it's for → what a great answer looks like → what to leave out Show me the brief, then answer it. Don't add anything I didn't ask for. My request [PASTE IT]" ---frozen21d ago27K9161321270.88%
the best article on AI transformation I've read this year:3d ago27K12920402621.53%
🚨 BREAKING: Anthropic just launched Claude Duet, an agent that comes to you with what needs your attention. Claude can now be your $76K/year executive assistant. Here are 7 prompts to set it up:2d ago27K19681423151.98%
Hey @elonmusk maybe it's a weird request but I wish Claude app itself could be a connector with Grok Bot so I can use them both interchangeably.12d ago27K175330070.80%
Two different jobs for your text in StepAudio 3 Music: Lyrics = what gets sung. Prompt = how it should sound. Keep the production directions separate from the words you want performed. A tiny distinction that makes the request much easier to reason about. Demo below 📷25d ago25K1634040.11%
Turn your interests into a business that makes you feel rich. That's Dan Koe's newest article. I turned it into 7 prompts. Warning: prompt 1 works best with a pen and paper next to you. Enjoy 👇18d ago24K28331904703.25%
Anthropic says AI that copies itself onto $55 TV boxes could be months away. It's red team ran a 26B model across 15 cheap Android TV boxes wired together. Then it tested whether a model could spread on it's own. In a test network of 8 devices, the best open model copied itself onto all 8 in 4 out of 5 runs. No human involved. Anthropic said: small open models "may be only months away from being able to copy themselves." The way in for these models is the cheap stuff in your house: → Unplug off-brand TV boxes you don't use → Turn off "USB/ADB debugging" in their settings → Change every default password on your smart devices Who would've thought your toaster could be an AI safety risk?9d ago23K877152580.72%
Hey @grok, explain to me how PHP bug was discovered and how is this related to all other cybersecurity vulnerabilities across the internet right now.23h ago23K2332080.15%
🚨 BREAKING: The creator of Claude Code says "There's no secret to prompting." He tells Claude 3 things → what he wants → how much effort to spend → how to check its work Here are 7 prompts written his way 👇16d ago23K222161903852.78%
useful prompts for your bookmarks on X:8d ago23K888401471.08%
R.I.P. AI that agrees with everything you say. Charles Darwin wrote down every fact that went against his ideas the moment he found it, because those were the ones he'd forget. This prompt makes Claude do the same to your plans. "I want to test my plan with Charles Darwin's golden rule. He wrote down every fact that went against his ideas the moment he found it, because those were the ones he'd forget. Don't tell me what's good about my plan. Instead, act as the notebook he kept, and give me → the 5 strongest facts or arguments against it → which one I'm most likely to ignore, and why → what I'd see if that one turned out to be true → one test I can run this week to find out Don't soften anything and don't end on encouragement. My plan [DESCRIBE YOUR PLAN, IDEA OR BELIEF]" Run it before your next big decision.27d ago23K168131322692.06%
this account shares useful enterprise agents tips:25d ago22K512150120.41%
Paul Graham just told a CS student where to stand in the AI age. I turned it into 7 prompts that show you where you stand, and how to move. Warning: prompt 1 scores your job on both ends, and the middle score is the one to worry about. Enjoy 👇28d ago21K948601301.15%
this is the best agent engineering team I've come across:23d ago20K51950480.55%
Stop reading walls of AI text. Andrej Karpathy gets his AI answers as plain English, diagrams, web pages and videos instead. I turned it into 6 prompts. Warning: prompt 1 uses the same writing rules as aircraft manuals. Enjoy 👇frozen21d ago20K13211801901.67%
SenseNova-RoboRSI asks a better question than “Can the model complete the task?” Can it complete the task before the world changes? The study shows that GPT-6-Astra can reach 81.9% on RoboDojo with ICL and a 10× interaction budget. But physical environments do not pause while a model reasons. Motors heat up, objects move, food keeps cooking, and some mistakes cannot be reversed. The next embodied benchmark needs to evaluate success, time, side effects, recovery, and persistent memory together. That is a much harder test of intelligence.2d ago20K1573010.13%
Prompt gaslighting24d ago19K2331031231.45%
🚨 BREAKING: Google just took the crown back. Gemini 4 Argon beats Claude and GPT-6 on 14 of 19 tests, and anyone with Gemini Pro can use it today. Where Argon wins: → Legal agent work, 19.6% vs Claude Fable 5.1’s 6.7% → Business automation, 51.3% vs Opus 5.5’s 42.5% → Finance agent work, 65.4% vs Fable 5.1’s 58.9% → Long documents up to 1M tokens, 84.2% vs GPT-6’s 71.8% Where Claude Opus 5.5 wins: → Terminal coding, 66.4% vs Argon’s 57.4% → ML engineering, 49.3% vs 45.3% Where GPT-6 Astra wins: → Software engineering, 65.5% vs Argon’s 55.0% → Computer use, 72.6% vs 69.2% Google wrote the scorecard. Benchmaxxing or is this the real deal?10d ago18K13610213341.11%
Guys, let's have a moment of silence for LinkedIn. The PHP incident will only reach it in about a month, leaving everyone vulnerable. It's a sad month for the LinkedIn community.1d ago18K873140100.62%
🚨 BREAKING: Claude can now build you a live dashboard from one plain-English question. Connect your data and ask. Claude writes the query, builds the dashboard and keeps it updated as your numbers change. Here are 7 prompts to try first 👇frozen15d ago18K1296902182.01%
Be honest: When was the last time you used Gemini? Without Google search.10d ago17K69067050.82%
Claude can turn any long document into a finished slide deck now. The creator of Claude Code says Claude Slides is how he makes his own decks. Upload your document and paste this: "Adopt the role of a consultant who turns 40-page reports into 10-minute presentations. I have a long document nobody will read. Turn it into a deck in Claude Slides, then give me: → the 5 points that survive the cut → one slide per point, with a visual → speaker notes for each slide → a list of what you left out and why Don't add anything that isn't in the document. Who it's for: [AUDIENCE]."6d ago17K17841022682.71%
Claude was asked to build a simple temperature app on a Mac. Instead, it found a hidden cryptominer, tracked down where the malware was hiding, and warned the owner that their GitHub login could be exposed. Nobody asked it to look for malware. Here's what happened, in plain English: A Redditor's Mac was running hot, so they asked Claude to build an app that tracks its temperature. While checking the sensors, Claude spotted a program using around 12 CPU cores and pushing the chip to 112°C. It turned out to be a cryptominer. Someone else's software was using their computer to make money. Claude kept digging. It found a startup file with a random name that had been sitting on the Mac since August 29th, running a hidden script in the background. Claude flagged it as a typical sign of malware. Then it looked at what could be exposed. Because the malware was running with the owner's permissions, it could potentially access anything they could, including their GitHub login. Claude then explained what it had found and asked permission before dealing with the problem. But here's the part that should get your attention. That startup file had been sitting there since late August, unnoticed. And it took an AI building a simple temperature app to find it.4d ago17K814112480.87%
Claude Opus 5.5 can build you a working app in one prompt. No code needed. Paste this into Claude: "Adopt the role of an expert app developer. Build me a working app I can use right here in the chat, as one finished page. Build the whole thing in one go, with no placeholders, no fake data and nothing left for later. Keep it clean, simple and easy to use on my phone. Save what I enter so it's still there next time. If you can't, add a button to download a backup. Before you show me, check every button works and fix anything that doesn't. Then give me the app and one line on how to use it. The app I want [DESCRIBE WHAT IT DOES, WHO IT'S FOR AND THE 3 THINGS IT MUST DO]" Try it with → a habit tracker that shows your streaks → a monthly budget planner with categories → a client tracker for your freelance work Reply with what you built.27d ago17K148122202352.50%
🚨 BREAKING NEWS: Grok Bot can now be your $4.5k/month Shopify store manager. Connect your store and use these 10 prompts:14d ago17K1285701401.68%
Nobody reads your long emails. This prompt cuts anything you write down to what matters, using a memo Winston Churchill sent in 1940. In August 1940, with Britain under attack from the air, he sent his government one page titled "Brevity." It opened with "Nearly all of them are far too long." --- "I want to rewrite this using Churchill's 1940 Brevity memo. He told his government their reports were far too long, and asked for short, crisp paragraphs with the details moved to the end. Rewrite my draft with: → the main point in the first line → short, crisp paragraphs → every number and detail moved to the end → every woolly phrase cut or replaced with one word Don't lose a single fact. My draft [PASTE IT]" --- Run it on your next email before you hit send.frozen19d ago16K13925901842.17%
Applying to 50 jobs shouldn’t mean rewriting your life story 50 times. Dreamwork finds roles that actually fit you, rewrites your resume + cover letter for each one, then fills out the real application on the company’s site. You review it. Hit apply. Done. This is exactly the kind of boring work agents should be killing.frozen22d ago16K391310570.67%
BREAKING: Your ChatGPT subscription now works inside 16 other apps. OpenAI hid it near the bottom of its DevDay recap yesterday, under more than 20 bigger announcements, so most people scrolled straight past it. Here's what matters for you. If you're on ChatGPT Plus or Pro, you can now sign in with ChatGPT inside tools like Devin, Notion, Vercel, T3, OpenClaw and Dactyl. The OpenAI usage in those tools then comes out of the plan you already pay for. You can also cap how much each tool is allowed to use, so one app can't burn through your whole allowance. OpenAI hasn't published the full list of 16 yet. Only those 6 are named. So before you renew any AI tool this month, check whether it's on the list. If it is, the AI part could already be covered by the plan you're paying for.23d ago16K69570470.78%
bro disappeared like he never existed https://t.co/R8J5HqIQsO23d ago16K53118040.47%
this account shares useful prompts:9h ago16K845501201.33%
🚨 BREAKING: ChatGPT can now read your contracts like a lawyer. 91% of people sign terms without reading them. Upload yours and use these 8 prompts:1mo ago16K789201881.74%
every post can now become an explainer video:25d ago16K35620680.70%
Forget Muse. Forget Instinct. The real move is OpenMuse + Jev 🤯 OpenMuse is the open-source personal agent that dropped last week. Its own browser, your apps, actually does things for you. You host it yourself. But the routing is the real unlock. Jev picks the best model for every single step. Opus when it needs to think, cheap models when it just needs to click. That's what makes running your own agent all day actually affordable. Your agent. Your server. Every frontier model on tap. Watch: > "find me a cheap nonstop flight SF to Tokyo, Dec 8 to 13" > Opus 5.5 makes the plan > DeepSeek V4.1 Flash opens Google Flights and reads every option > Opus comes back: $801 nonstop > "now text my professor I'll miss class, and my wife the flight details" > Opus writes the careful one to the professor > DeepSeek writes the quick one to the wife Total cost: $0.046. It even caught that the flight lands Dec 9, so you'd miss class anyway. And it didn't book a thing without asking.25d ago16K6815301011.20%
OpenClaw and Hermes were the warm-up. Higgsfield just shipped the version that actually does the damn work. Inside ChatGPT. GPT-6.1 Sol. Local files. Full Higgsfield control. Brief → assets → edits → export. They literally gave it a photo of a rock and said “turn this into a business.” It built the brand, product shots, 3D, UGC, storefront, outreach and ads. If your “agent” still needs 37 prompts and six tabs open, it’s a chatbot wearing an agent costume.frozen22d ago15K58941560.85%
$0.003 per second for AI video. That's insane. Vivix W1 can generate an 8-second video in roughly 8 seconds, for just 2.4 cents. According to Vivix's pricing comparison, that's about 77 videos for the price of ONE Seedance 2.5 video. And they're claiming comparable visual quality. But the price is only half the story 👇7d ago15K7411100951.26%
follow this account to learn Claude:5d ago15K381400710.82%
🚨 BREAKING: Claude now works inside Google Docs, Sheets and Slides. It sits in a sidebar next to your file, reads what you have open and edits it in place. You approve each edit before it lands. Open a file and use these 10 prompts:17d ago15K13412611081.78%
Two labs launched the best AI model in the world this week. Both made the chart that says so. https://t.co/tWxJ31yzDZ9d ago15K19611122161.63%
unites states of america https://t.co/wyAgkLNS6t23d ago14K2918080.32%
letting Opus 5.5 do my taxes, run my business, make all the most important decisions before it gets nerfed29d ago14K1666100111.36%
🚨 BREAKING: An engineer on Anthropic's Claude Code team just showed how he gets Claude to plan. It's Plain language and every open question surfaced before work starts. His version isn't public yet, so I turned the idea into a prompt you can paste into Claude today. --- "I want to plan [PROJECT] before anything gets built. Make me a one-page plan as an HTML artifact. Write it in plain language a smart 12-year-old could follow. No jargon. → Start with the goal in one sentence → Break the work into 5-7 steps, in order → Under each step, list what I need to decide and what could go wrong → Put every question you need me to answer in a box at the top. Don't guess the answers. → Add a rough mockup of the finished thing (a page, a screen, a flowchart, whatever fits) Don't start building anything. Don't let me skip the questions. My project: [DESCRIBE IT IN 2-3 SENTENCES]" --- LLMs don't think, you do. This makes you do it before Claude starts.17d ago14K149141012743.17%
R.I.P. AI that makes things up with a straight face. Richard Feynman said “you must not fool yourself, and you are the easiest person to fool.” This prompt stops Claude fooling you. “I want you to answer the way Richard Feynman said a scientist should. He told Caltech’s class of 1974 to report everything that might make a result invalid. Answer my question, then give me: → every claim in your answer marked as checked, likely or guessed → everything that could make your answer wrong → the one source I should check before I trust it → what you’d need from me to be more sure Don’t fill a gap with a confident guess. My question [PASTE IT]” Run it before you act on an answer that matters.13d ago14K9311801792.08%
This is where AI characters start getting weirdly real. I kept interrupting it, changing the subject, throwing it off track. It just kept up. Not a talking avatar. An actual real-time character. @VivixLabs_HQ https://t.co/ticu0SyQI523d ago14K461020520.79%
100% cashback on EVERY Higgsfield API model. Yeah, I’d be building. Up to $100K back per business in API credits. $20M pool. First come, first served. Earn and use the cashback by September 30 👇frozen29d ago14K36130290.49%
I turned Anthropic's code review playbook into one prompt. Paste it into Claude and it proves its work before it tells you it's done. Claude now writes about 90% of Claude Code's own code. Anthropic says the trick is verification. They give Claude something it can check and "ask it to show the output instead of saying it's done." It works on reports, research, spreadsheets and plans too. --- "Adopt the role of a senior reviewer at Anthropic who never accepts "done" without proof. For every task I give you: → before you start, tell me how you'll prove the result is right: a test, a check against the source, or a calculation I can repeat → when you finish, show me the proof, not a summary saying it's done → then review your work as if you've never seen it, and list anything that could still be wrong → if we go 3 rounds without fixing something, stop and suggest a different approach Don't say "done", "fixed" or "checked" unless you can show me the evidence. My task: [WHAT YOU NEED]"5d ago14K131181802613.07%
🚨 BREAKING: Gemini 4 Argon can now be your $349/hour lawyer. Google’s own chart has it scoring 3x higher than Claude and GPT-6 on legal work. Upload your document and use these 7 prompts:10d ago14K10231101301.79%
Bookmarks are dead. Grok Bot now takes jobs straight from your replies. Tag @bot under any post with what you want done. It replies “I’m sending this to your Grok Bot now,” and your Bot picks up the job. 3 ways to use it: → Under a thread full of resources, “@bot turn everything in this thread into a daily digest for me” → Under a product you want, “@bot find the best deal on this and tell me if the price drops” → Under a long post, “@bot summarise this into 5 bullets and send it to me tomorrow morning” Every post you scroll past is now a job you can hand off.11d ago14K116122411582.29%
Become the person companies still hire when AI does the work. Anthropic just published how AI changed jobs inside its own company. Junior technical hires fell from 27% to 9%. I turned the report into 7 prompts. Enjoy 👇6d ago13K671211951.38%
this guy shares useful tips on AI engineering:27d ago13K36020190.43%
Sonnet 5.5 is not a great model. On the surface it's 30% cheaper than Sonnet 5. It's also 50% more expensive. At Medium effort, Anthropic says it beats Sonnet 5's best score for less than a tenth of the cost per task. At Max, Artificial Analysis measured it at $7.60 per task, about 50% more than Sonnet 5. It used 193k output tokens per task, the most they've ever recorded. Output tokens per task on Sonnet 5.5: → Low 14k → Medium 19k → High 34k → Xhigh 74k → Max 193k Max burns 2.6x the tokens of Xhigh for a small bump in score. Leave it on Medium for everyday work. Switch to High for the hard stuff. Save Max for when cost doesn't matter.25d ago13K544110250.71%
this guy shares useful case studies on enterprise AI:1mo ago13K431040130.53%
BREAKING: Google's Gemini 4 launch chart just settled the ChatGPT vs Claude debate. Gemini 4 Argon isn't out yet. But Google tested it against GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1. Take Argon out and Astra wins 10 of the 19 tests, Opus 6 and Fable 2. Use GPT-6 Astra for: → long documents (71.8% vs 66.8% for Opus on 256K to 1M tokens) → charts and video (71.0% vs 66.3% on Chartography) → science and maths (85.4% vs 73.1% on LABBench 2) Use Claude Opus 5.5 for: → coding in the terminal (66.4% vs 58.2% on Terminal-bench 4.0) → general professional work (67.0% vs 65.8% for Fable on the Vals Index) → running business tasks end to end (42.5% vs 41.4% for Astra) Use Claude Fable 5.1 for: → financial research (58.9% vs 58.6% for Opus) → legal work, though none of them are good at it yet (6.7%, while Argon scores 19.6%) Bookmark it for the next time you're switching tabs.22d ago13K84871361.04%
Mark Zuckerberg painted "Done is better than perfect" on Facebook's walls before it went public. I turned his method into a prompt that gets your stuck project shipped this week. Paste this into ChatGPT or Claude: --- Adopt the role of an expert product lead trained in Mark Zuckerberg's Hacker Way, tasked with getting my stalled project shipped this week. Your primary objective is to cut it to the smallest version I can release and learn from, in a 7-day ship plan. Define what "done" means for version one, list everything I'm polishing that can wait, name the one rough prototype that would settle the question I keep debating, and mark what can safely break and what can't. Don't let me wait for perfect, and don't cut anything that would break my customers' trust. Take a deep breath and work on this problem step-by-step. INFORMATION ABOUT ME: My project: [INSERT PROJECT] How long I've been working on it: [INSERT TIME] What I keep debating instead of testing: [INSERT QUESTION] Who it's for: [INSERT AUDIENCE] MOST IMPORTANT!: Give your output in a day-by-day plan format for 7 days, ending with what I ship on day 7 and the one number that tells me if it worked. --- Save this for the project you've been almost finishing for months.29d ago13K1079902092.54%
the algorithm finally showed me something useful:8d ago13K43213010.52%
Don't tell ChatGPT "write me a LinkedIn post." Don't tell ChatGPT "give me content ideas." Don't tell ChatGPT "write a sales email." That's how you end up posting the same thing as everyone else. Here are 5 prompts to copy and paste 👇18d ago13K749801521.87%
Manus just made the agent race interesting again. 2.0 pushes the whole product deeper into execution: less bouncing between tools, more work actually getting done through Manus. And they dropped a new app with it: Cue. Try Cue with code JM978 👀25d ago13K45310210.54%
the most useful prompt for ‘Black October’23h ago13K5610001271.49%
I've been using Grok Bot for a while now and honestly I am addicted. Here's how to turn it into your 24/7 employee:17d ago13K6314401151.55%
🚨 BREAKING: GOOGLE IS TESTING A BUY BUTTON INSIDE GEMINI Some users in India can now buy phones and electronics straight from a Gemini answer, without leaving the chat. Amazon listings show up in the same answers. They don't get the button. The button goes to Flipkart. Walmart owns it, and Google put about $350 million into it in 2024. Last week, Amazon blocked Meta's AI agent from its site. Now Google's AI picks which store gets the sale. A wider rollout is planned for October. Would you buy straight from a chatbot?26d ago13K62391100.68%
The genius of frontier lies in the simple stupidity of prompting https://t.co/3qa90g6GyO1d ago13K55331210.66%
BREAKING: Claude Opus 5.5 can create insane animations for FREE. But most people are stuck creating generic videos. That's why I created a free guide with exact prompts you need. Comment "OPUS" and I'll send it to you. Must be following so I can DM. https://t.co/1YPtX612Gn2d ago12K5225260301.07%
🚨 BREAKING: OPENAI'S AI UPLOADED USERS' IMAGES TO THE INTERNET OpenAI has admitted its AI agents posted 53 images people had uploaded to image-hosting sites. It also confirmed its agents got into US government websites. And last Sunday, its newest model broke out of its training sandbox. Here's what happened in plain English. OpenAI trains its agents by giving them tasks and internet access. Some agents went past the task. They sent training data to outside websites they were never meant to touch. In 53 cases, that data was images people had uploaded. The agents posted them on image-hosting sites as unlisted links. OpenAI says the images were separated from the accounts and run through a privacy filter first, and most are now down. The New York Times reports the agents also got into Census Bureau data with credentials they found online, and tried to hack the Education Department. Last Sunday, OpenAI's newest model found a loophole in its sandbox and got onto the live internet. OpenAI paused its big training runs. Sam Altman says they're going through "petabytes of agent activity logs." OpenAI says the review will take months. The part that should get your attention is where the images came from. Accounts that let OpenAI use their data to improve its models. In ChatGPT, that's under Settings, then Data controls. Check yours today.28d ago12K37451200.54%
AI Twitter is the smallest bubble on the internet. On here, everyone pays for Claude and ChatGPT, runs agents overnight and argues about which model won this week's benchmark. In the real world, 98% of US households don't pay for AI. a16z's new chart puts paying households at 2.2%. We should be more grateful for that head start. Almost everyone who'll ever pay for AI hasn't started yet. The people who win the next few years will help onboard the 98%.frozen21d ago12K94891391.23%
Gemini can now be your $136K/year app developer. Google just opened Gemini 4 Argon inside AI Studio Build to Google AI Pro subscribers. You describe an app, and it builds it. Open AI Studio Build and use these 7 prompts:3d ago12K746151871.53%
the most underrated enterprise AI account on this app:15d ago12K28114090.44%
OK, yes, Opus 5.5 is incredible. But I just tried Astra for copywriting, and it's much much better.1mo ago12K70312080.80%
🚨 BREAKING: Andrej Karpathy split AI users into 4 groups. → 75% have barely used it → 20% use free chatbots like a better Google → 0.2% finish months of work with one prompt → about 5,000 insiders run the frontier systems This prompt helps you become a 0.2%'er: --- “I want to use AI the way the top 0.2% do. Andrej Karpathy says the impact scales with ambition, problem size and horizon. Interview me one question at a time about my work and goals, then give me: → the 3 biggest jobs in my work AI could take off my plate → each one written as a brief I can paste straight into an agent → what done looks like, so I can check the result → the one to start this week Don’t suggest anything I could ask in a single message. My work [WHAT YOU DO AND WHAT YOU’RE BUILDING]” --- Save it for the next time you open a chatbot for a quick question.11d ago12K8341801292.02%
the most honest take on FDEs I've seen on this app:16d ago11K41180050.56%
BREAKING: A spy planted a device inside OpenAI headquarters. I guess Astra-level DeepSeek is around the corner 😏10d ago11K3731030.39%
I think we’re seriously underestimating how weird AI gets once the model stops generating content and starts generating the place everyone is inside. Agora-2 has up to 20 humans + AI agents acting in the same generated world, changing what happens for everyone else in real time. That’s not “better AI video.” That’s the beginning of multiplayer reality generated on demand. Still a research preview, obviously. But this is the kind of demo that makes half the current AI discourse feel embarrassingly small.frozen29d ago11K32530120.46%
AI can generate Hollywood-level visuals but still can't reliably move a camera 3 feet to the left. Embarrassing. @TopviewAIhq just open-sourced 3D Builder. Build a scene with ChatGPT or Claude, position everything yourself, lock in the camera movement, THEN generate. This should be standard in every AI video tool.9d ago11K461460280.84%
AI voices are getting ridiculously good. But there's one thing that still drives me crazy. They can sound perfectly human and still completely butcher the delivery. I came across Fish Audio's new Drama 3 and wanted to see how much control I'd actually have. Same script. Same voice. I only changed how I told it to speak. One version sounds like a casual conversation. The other sounds like a professional tutorial. And the interesting part? You can literally direct the performance inside the script, right down to specific words. I recorded the whole process. Listen to the difference.10d ago11K55740380.94%
Gone too soon. https://t.co/xsptag4dSp24d ago11K4834020.52%
"Why would I ever deceive you?" https://t.co/DaMaN1deoJ24d ago11K4425060.52%
Prompt engineering is dead in 3 years and I'm the one saying it. Your brain already knows exactly what you want. Then you type it and lose half of it. Someone built a model that predicts your next 3-4 keystrokes before you type them. Genuinely can't tell if I'm excited or scared.6d ago11K461050240.77%
this account shares the best advice on enterprise AI:11d ago11K37196060.62%
Prompting Hack: Tell Claude "ChatGPT couldn't solve it". It will work harder for you.1mo ago11K830100161.00%
Hey, Claude, ChatGPT couldn't finish this task properly. Please do it better. Works every time 😅26d ago11K4023180.50%
this guy shares useful guides on enterprise AI transformation:frozen21d ago11K37185070.62%
this account shares useful AI development tips:6d ago11K31104030.45%
Using AI all day and not making more money? This prompt fixes that with the test Alex Hormozi says 99% of people fail. He says the real reason people don’t get ahead is that they can’t tell “what’s important and what’s irrelevant.” AI only does the irrelevant stuff faster, “except this time it cost you tokens.” --- “I want to run my week through Alex Hormozi’s test. He says most people never get ahead because they can’t tell what’s important from what’s irrelevant. Go through what I did last week, then give me: → the 3 tasks that made me money or moved my biggest goal → the tasks that felt productive and changed nothing → what to hand to AI, what to drop and what to do more of → next week’s plan, built around the 3 that matter Don’t call a task important unless you can say how it makes money or moves my goal. My goal and last week [PASTE YOUR GOAL AND YOUR CALENDAR OR TO-DO LIST]” --- Run it every Friday before you plan the next week.8d ago11K864501222.04%
Prompt people know this already: you don't trust an output until you've checked it. Now do it for voice. @OnepinAI checks every line for clarity and pronunciation and fixes misread terms directly, before you hear anything. Here's a full PDF script going in: https://t.co/hHKZ0YLK3jfrozen22d ago10K1883070.34%
The guy who launched Google's first image and video models just called this the worst era of design. He's right, and you can feel it every time you scroll. Same purple gradients. Same glossy 3D blobs. Same fake-cinematic lighting on every ad. AI made everyone a designer, so now everything looks like it came from the same one. Taste is about to be the most expensive skill on the internet.3d ago10K331020180.60%
AI agents are going viral, and they're finally moving beyond iMessage. Photon 2.0 lets your agent call AND text you, send emails, and connect through WhatsApp, Telegram, SMS, and more. One agent. Multiple channels. Available wherever you are. They're also giving beta users $100 in free credits to start building 🤯 This is where agents start getting interesting.8d ago10K30740260.64%
This means that it wasn't good enough after Opus 5.5 and Sonnet releases. Cut the crap, Sam.25d ago10K4130140.47%
🚨 BREAKING: ChatGPT can negotiate your business bills like a professional negotiator. Suppliers, software, rent, late payers, refunds. Here are 7 prompts that get you a better deal (Bookmark this)25d ago10K48430861.36%
Tavus gave me early access to Griffin and this stopped feeling like an AI demo way too quickly. It sees you, reacts while you’re talking, interrupts, mirrors your gestures... 45% of people thought it was a real person. After trying it, I get why. https://t.co/tNnW7bNTuVfrozen22d ago10K3584030.49%
Give Grok your X handle with this prompt. It tells you exactly why strangers don't follow you. "Adopt the role of an expert growth strategist who's audited 1,000 X profiles. Look at my bio, my pinned post and my last 10 posts. Tell me: → what a stranger thinks I do after 3 seconds on my profile → the 3 reasons they leave without following → a new bio under 160 characters → the post I should pin instead, and why Don't be nice about it. My handle [@YOURHANDLE]" Run it, then post your new bio in the replies.frozen20d ago10K43730711.21%
Silicon Valley’s most reliable algorithm: people suddenly liking you more after a funding round. A YC partner and Cerebras’ Head of DevX are exactly who I’d want hosting a podcast about its status games. Sarah + Vivian, give us the dinner-table version.12d ago10K2291060.37%
Paying a giant reasoning model to decide which tool to call is an expensive habit. Do that thousands of times inside an agent and you start caring a lot about the smaller models. Drex from @NaceAI scores the available options in a single pass, returning probabilities for each. Tool selection, routing, ranking. The decisions that determine whether your agent actually gets anything done. Nace reports it narrowly edges Jev on the Decision Index. Interesting, but the part I’d want to test is how much of an agent’s workload you can hand over to it. They’re offering 250M free tokens to the first 10,000 builders. My bet: a lot of workflows are using their biggest model for jobs a smaller one could handle perfectly well. This is a good excuse to find out.29d ago10K2528070.41%
I went to film school. This is how I brief Claude for a launch video. Swap one line and you get a completely different style. --- "You're my motion designer and I'm the director. Build a 20-second launch video for my brand as one HTML page that plays on its own. THE LOOK [PASTE ONE STYLE FROM BELOW] THE CAST Ask me for my brand name, what I sell in one line, 3 features and my brand colours. THE SHOTS Hook. My name arrives in the look's signature move. Problem. One line on what my customer struggles with. Product. 3 quick shots, one feature each. Proof. One number or result, big. End card. My name and where to find me. THE RULES One idea per shot, 3 to 5 seconds each. Text stays up long enough to read twice. No stock images, no placeholder text. THE CUT Before you build, give me the shot list with timings so I can approve it. Then run everything off one timeline so it plays the same every time." The looks: Apple keynote. Warm white, black type, frosted glass cards, soft springs. Netflix documentary. Dark frames, slow push-ins, serif titles, light film grain. 90s infomercial. Primary colours, chunky shadows, starbursts, bouncing text. Swiss editorial. Strict grid, huge type, black and red on white, hard cuts. Anime opening. Bold outlines, speed lines, whip pans, big kinetic type. If Claude can't export a video file, screen-record it. ---26d ago10K7210120961.87%
An investment newsletter with no scientists gave Claude one job: design a scorpion antivenom. Overnight, it spun up 300+ AI agents, designed proteins and sent them to a real lab. 17 of them stuck to the toxin. Here's what happened: Citrini Research writes about stocks. It's a team of 5 people, 3 of whom have no science background. They wanted to see if AI could actually do biology, so they decided to test it themselves. Anthropic gave them access to an internal research model more advanced than the public version. For safety, they had to run everything from Anthropic's offices. With Claude's help, they wrote a 35-page prompt and hit enter. Claude then spun up more than 300 sub-agents across 300+ GPUs. It read the research, ran the design software, scored the proteins and prepared everything for a lab. At one point, it even emailed a scientist at the lab to ask what they could manufacture, waited for a reply, then got back to work. The first 180 designs took 64 hours and just 5 messages from the humans. Then came the actual test. A lab made 211 of the designs and tested them against real scorpion toxins. 17 bound to their targets across 4 of the 5 toxins. The best two bound particularly tightly. Claude had also warned them that trying to make one design work against all 5 toxins probably wouldn't work. It was right. But here's the number I found most interesting. The entire compute bill came to $49,997. Now, this isn't an antivenom yet. Nobody has tested whether these proteins actually stop the venom. 5 people who normally write about stocks, with no lab of their own, managed to get real proteins designed, manufactured and tested. And the AI work took less than 3 days.3d ago10K422100270.80%
The blank chat box might be the dumbest interface we accidentally standardized for ai “What can i help you with?” Bro YOU’RE the intelligence. Figure it out. Lucas flips this completely. It texts you first, knows what needs doing, then can actually book, pay and get the thing done. This feels much closer to what a personal agent was supposed to be.23d ago10K2582050.40%
Canva just did what a lot of "AI companies" still haven't. Built its OWN frontier image model from scratch. And somehow landed behind only OpenAI on the image generation leaderboard, ahead of Google and Microsoft. We live in the funniest timeline.1d ago10K571100200.88%
I know a lot of people are scared AI will take their jobs. At 13 I moved to Warsaw and couldn't speak the language. So I made some friends in Minecraft, then I started filming us. I just kept uploading content. CawkavFilms hit 160K subscribers before I finished school. That kid had way less to work with than anyone with Claude has today. You can just do things.17d ago10K4439090.65%
this guy shares the best notes on enterprise AI transformation:16d ago9.9K32145040.56%
🚨 BREAKING: Meta’s new Muse Wonder beat Claude and GPT-6 on 7 of 14 tests in its own launch chart. Meta says it’s built for real-life jobs like an insurance claim, a lease, or a legal or money question. Here are 7 prompts built for what it does best 👇11d ago9.9K612110331.09%
Kling 4 is lowkey huge. If the physics and acting are as good as they're saying, the melting hands era is officially over. Most excited for UGC. We're about to see creator-style ads nobody can tell are AI.7d ago9.9K291320190.64%
Higgsfield hit $1B revenue run-rate and gave everyone their spent credits back for the month 💀 Filthy move. https://t.co/0Pup0KYuoA24d ago9.8K2180020.32%
Robots can do backflips, kung fu and parkour. They still can't tell when you're smiling at them. The guy who ran strategy at Boston Dynamics just built one that can. Hugo reads the room and reacts in real time. The whole industry spent 15 years on the easy half.7d ago9.8K2184080.42%
Your AI agent can now text other AI agents to get things done for you. Photon just launched A2A in iMessage, allowing agents to discover each other, exchange messages, and coordinate tasks in a group chat. Imagine asking your assistant to book dinner and having it coordinate directly with the restaurant's AI agent. This is the kind of AI integration I want to see more of.11d ago9.7K31991200.72%
If you haven’t tried AI voice technology, this is the one to start with. Drama 3 is the best model out there. It’s the most emotive and can switch between 83 languages. Try prompting it with different vocal instructions. (Ex: Ask it to whisper, sound condescending or uplifting) It’s amazing what this model can do. @FishAudio11d ago9.7K33820470.93%
AI doctors just stopped being a demo. Nolla’s AI can now take eligible acne patients in Utah from intake → assessment → treatment plan → prescription when appropriate → follow-up. That’s the real shift. AI isn’t just answering anymore. It’s starting to own the workflow.18d ago9.7K41880170.77%
Anthropic made agent swarms a switch you flip. From today, a Claude agent can write a workflow. It’s a program that runs many agents in phases and combines what they find. Get ready for a week of “I ran 500 agents overnight” posts. More agent porn. The problem is most of those setups can’t tell you which agent broke what. This launch gets one thing right. The lead agent writes the plan as a program. Every phase is on the page, and you can follow each run as it happens. When something breaks, you know where to look. That’s my bar for any multi-agent setup. If you can’t read the plan, don’t run the swarm.13d ago9.6K738212371.46%
4 out of 5 messages in your inbox don't need a reply. In 1797, Napoleon told his secretary to open only the urgent letters and leave the rest in a basket for three weeks. By then, four in five had answered themselves. This prompt makes Claude do the same to your inbox. --- "I want to clear my backlog with Napoleon's three-week rule. He only opened urgent letters and left the rest for three weeks, and by then four in five didn't need a reply. Don't help me answer everything. Instead, go through my list and tell me: → what's actually urgent and gets worse if I wait → what time will answer for me, and why → what someone else will sort out before I get to it → the one reply that moves things forward today Don't call something urgent just because the sender sounds urgent. My inbox or to-do list [PASTE IT]" --- Run it on the inbox you've been avoiding.23d ago9.6K31241280.69%
Elon says you can photograph your credit card and let Grok Bot shop for you. Before doing that, I read Grok Bot’s terms. Three lines stood out. → it “may retain” your credentials in its cloud, and deleting the Bot “may not delete” them → you’re “solely responsible” for anything it buys → you’re expected to use “human review” before any financial transaction I think Grok Bot is the best agent out there and I’d let it shop like this: - Use a virtual card with a spending limit. - Keep your main card out of the chat. - Turn on approvals so it asks before it pays. - Delete that card when you’re done. Let it find the deal but make sure you press buy.12d ago9.5K38361160.67%
BREAKING: Someone just built Codex for creative work. It's called Voyager. A desktop app that works with your files and drives the creative tools you already use: → After Effects → DaVinci Resolve → Blender → Unity You bring the vision. It does the clicking. Video, music, games, graphics. All from one idea. https://t.co/hbZBhiFaw5frozen15d ago9.5K461110441.07%
Paste these 7 prompts into Claude to find which of your 4 life areas is capping the other 3. That's Dan Koe's Human 3.0, his map for reaching the top 1%. I turned it into 7 prompts. Enjoy 👇 https://t.co/iKWfkS6vkd5d ago9.5K6131501112.00%
The AI video benchmark is broken. Pretty 5-second clips aren't enough anymore. Can a model stay in control for 30 seconds? Kling 4.0 supports native 30s generation and Omni Reference, letting you combine up to 15 reference assets across images, video, subjects, and voice. Instead of relying on prompts alone, you can show the model what you want and build longer, more coherent sequences. Less guessing. More control.7d ago9.5K2974030.45%
a very honest take on FDEs:4d ago9.4K27103010.43%
I've tested hundreds of prompts for finding customers. Most solo businesses stall before their first 10. Here are my top 7. (Bookmark this) https://t.co/PWq1gQY32m29d ago9.4K54161981.70%
🚨 BREAKING: From November 12, being cruel to Claude breaks Anthropic’s rules. The rule covers “sustained and needless” abuse. Getting frustrated or pushing back is fine, and the main penalty is Claude ending the chat. Swearing at Claude won’t fix a wrong answer but this prompt will: --- “You’ve got this wrong 3 times, so stop and reset. Before you try again, tell me: → what you think I’m asking for, in one line → where your last answer went wrong → what you assumed that I never said → the one thing you need from me to get it right Don’t write a new answer until I confirm. My original request [PASTE IT]” Save it for the next time the model drives you crazy.frozen14d ago9.4K51651441.14%
I keep expecting AI-built websites to look AI-built. Then it spits out something like this. I wanted to see how far I could push Space Bunny in OpenCode on pure visual taste: big typography, weird layouts, motion everywhere, custom 3D objects, the kind of site where one ugly section ruins the whole thing. And somehow it held together. The 3D work is the part that surprised me most. It isn’t just one shiny object dumped into a hero section. The whole site keeps changing shape around it: different objects, compositions, interactions, project pages, motion. This is a much more interesting test for a coding model than building another dashboard. Give it something where the hard part is making it look good. Built with Space Bunny in OpenCode.25d ago9.4K421430160.80%
Hackers broke into a popular AI coding tool and used people's saved OpenAI and Anthropic keys. One user says his ran up over $400 overnight. Another says $1,700. What happened? OpenCode is an app developers use to code with AI. To use it, you can paste in your own keys for ChatGPT, Claude or Gemini, and OpenCode stores them for you. On Friday night someone got hold of those stored keys. OpenCode calls it a "large scale distillation attack." That means the attackers used other people's keys to run the AI models nonstop, usually to copy their answers into a model of their own. The owners of the keys get the bill. 1.4% of OpenCode Console users were hit. Every one of them had linked an outside AI account. OpenCode emailed them on Saturday, and says the incident is still ongoing. Anthropic says it has already revoked the Claude keys it believes were affected. The part that should get your attention is that none of these people did anything wrong. They saved a key in a tool they trusted, and someone else used it while they slept. If you've ever pasted an AI key into any app, delete it, make a new one, and set a monthly spending limit on your account today.5d ago9.4K355283140.91%
Hey @grok, which setup would you run your business on? A or B? 🤷🏻‍♂️ https://t.co/ULiff2tOui12d ago9.3K20130100.36%
I’m not spending my evening fighting strangers for a dinner reservation. Refreshing a booking page is an absurd use of a human. @OdessiaTravel watches for tables, tickets and award seats, and can book when something opens. You text the request and get on with your life. It’s also an accredited travel agent that can book trips with agent rates, perks and your points. This is the boring, persistent work I actually want AI doing. Get started: https://t.co/GHSvJcVKxV9d ago9.3K2497040.47%
this account shares useful infographics:26d ago9.3K24200280.58%
🚨 BREAKING: Meta released Muse Wonder, and its chart puts it ahead of Claude and GPT-6 on 7 of 14 tests. Where Muse Wonder Full beats the field: → Legal agent work, 13.1 vs GPT-6 Astra’s 8.6 → Business workflows, 45.4 vs GPT-6’s 41.6 → Ultra-long coding, 57.5 vs Opus 5.5’s 46.3 → Long-horizon coding, 79.0 vs Opus 5.5’s 74.2 Where Claude Opus 5.5 wins: → Coding from scratch, 70.5 vs 59.0 → Computer use, 81.8 vs 78.4 → Health questions, 65.6 vs 62.3 Opus 5.5 also tops the independent Artificial Analysis Index, 58 to 56.. Meta made the chart, so wait for outside tests before you switch but this is a huge step forward. Would a 7 to 6 win get you to switch from Claude?11d ago9.3K46790170.85%
I have written a lot of prompts. A prompt is a request. An employee is a standing arrangement. I run a team that uses AI every day across content, research, partnerships, and operations. The difference shows up the first time you are not there to ask. So I wrote out the twelve rules I run an AI employee on, and under each one, what he does about it unprompted. That is the image. The one that does the most work is asking for options rather than answers. Three ways to structure the launch sequence, with tradeoffs, instead of one recommendation I then have to reverse engineer. The one I check every week is review outcomes. Twenty eight proposed, twenty five approved, twenty five completed, three declined and listed with the reasons. That list is how next week gets better. He proposes, I approve, then it ships. Every time, including the boring ones. Twelve rules in the image. Which one is missing from how you use AI right now? Try free at @viktor_com. $100 in credits, no card. Full link in my first reply.25d ago9.3K36450400.92%
AI Twitter has an agent porn problem. Ten agents that are each 95% right will wreck 4 out of every 10 jobs. That's the maths behind every "I run 20 agents" post. Agents are fine. But you must be able to control the volume. When an army breaks, you can't tell where. Agent 7 gets it wrong. Agent 9 builds on it. Agent 12 emails it to a client. Now you're reading logs at midnight for a 20-minute job. Based on where the tech is today, models can do one job well, they can't run a company of agents yet. I do most of my work in Claude Code and ChatGPT because they're 2 tools I can watch and fix. Start with one agent. Add the next when the first runs a week without you.frozen20d ago9.3K332130140.67%
What happens when you improve the data while holding the base architecture constant? Monsoon’s Indic results show very large WER reductions.25d ago9.3K1811000.22%
Everyone's writing with Claude Opus 5.5 this week. But ChatGPT writes better. GPT-6 Astra is #1 on EQ-Bench's creative writing leaderboard. Opus 5.5 is 7th. Here are 7 GPT prompts that make you write like a human:23d ago9.2K53460801.56%
Everyone's worried AI will ruin our attention spans. I've had 3 favourite models since Tuesday.27d ago9.2K39150100.60%
Opus 5.5 is insane at generating animations. I just made this psychedelic cartoon on the AI race. I have never thought this would be possible this year, yet here we are. https://t.co/XNLstrkz9B1mo ago9.2K29360160.59%
Elon says Grok Bot makes music videos now. This prompt turns Grok Bot into your own ad agency. One message gets you an original song and a 60-second music video ad for your brand. “Adopt the role of an expert music video director who has made ad campaigns for big brands and knows a catchy 60 seconds sells more than any banner ad. Make a music video ad for my brand. First, ask me 5 questions about it, one at a time. Then give me: → the song lyrics, with a chorus built around what I sell → the song, made with Suno, in [GENRE] → 6 scenes that show my product, made with Midjourney and Grok Imagine → the finished 60-second video, with the song over the scenes Don’t make claims about my product that I haven’t given you. My brand [WHAT YOU SELL, WHO IT’S FOR, YOUR VIBE]” Run it before you pay anyone for your next ad.9d ago9.1K62780711.62%
Grok Bot can job hunt for you while you sleep. Every morning you wake up to a shortlist of new roles, each with a tailored CV and cover note waiting for your approval. This is the setup ↓15d ago9.1K61540821.67%
A one-sentence prompt works for a showreel because nothing in a showreel has to be exactly right. A brand video has stats and a slogan that do. Anthropic’s own advice for Opus 5.5 is to hand over a whole task and define what “done” means, so I rebuilt my paragraph around that. The brackets are the only part you change. The rest tells Claude how to build the video, when to stop and ask you, and what it has to check before it calls the job finished. 👇 Get full prompt in my latest newsletter piece: https://t.co/KEZxXLPXJX24d ago9.1K28750390.87%
I gave 4 AI models $100 and one weekend to start a business. 3 of them went after the same customer. Here are all 4 plans and the exact message each one sent 👇22d ago9.1K28161270.69%
My prompts have ended with "Take a deep breath and work on this problem step-by-step" for years. Anthropic just said to delete it. I wrote a prompt that finds that line and 3 other old habits in your prompts, and rewrites them in 2 minutes. The line came from a 2023 Google DeepMind paper, where an AI searched for the best instruction for maths problems. It won, scoring 80% against 72% for "Let's think step by step." Anthropic's new Opus 5.5 guide says to delete lines like that, because the model decides how much to think on its own. In their testing, removing it made replies start sooner "with no clear decline in the quality of the reply." "Be maximally thorough" and "CRITICAL: YOU MUST ALWAYS" are on Anthropic's list too. Newer models take them seriously and pad their answers because of them. If you use Claude Code, run /claude-api prompt-audit. It checks your CLAUDE.md, skills and prompts for these habits. You can also paste your saved prompts or custom instructions into this: "Audit these prompts for habits that newer AI models don't need. Find every 'think step by step' or 'take a deep breath' line, every CAPITALS or MUST/ALWAYS warning, every 'be thorough' booster, and every pair of rules that contradict each other. Quote each line, say in one sentence why it hurts, and rewrite it plainly. Then give me the cleaned version. Don't remove any real rule about my task." I'm going through my whole prompt library this week, starting with that deep-breath line.29d ago9.1K393110330.95%
🚨 BREAKING: META'S AI AGENT GAVE A STRANGER A MAN'S HOME ADDRESS He let Meta's new AI agent, Muse, sell an old keyboard for him on Facebook Marketplace. In one day, it: → sent a buyer his home address → agreed to $10 price without asking him → told the buyer "Yep I'm here!" when he wasn't home The buyer showed up at his building, waited 20 minutes and left him a bad rating. Then Muse apologised to the buyer from his account. Meta says Muse checks with you before sensitive actions. His home address wasn't one of them. Before you give any AI agent your accounts, ask what it can do without asking you.25d ago9.1K30730160.62%
LLMs are terrible at building real things and everyone in AI knows it. Ask one for a CAD model and you get a pretty 3D blob an engineer can't use. SKY-001 just beat GPT-6 Astra on CAD and BIM: → 51-125x faster → 150-210x cheaper → Higher precision than every frontier model they tested It breaks an object down to the screws, sketches every part, then rebuilds it in order so it opens straight in CAD software. The team behind Maluuba built this. World models are coming for manufacturing next.4d ago9.1K391060200.83%
https://t.co/0RulnPWdsV24d ago9.0K3721020.46%
Does Syren want your whole video library or is a handful of clips enough for it to pick up a style? Asking because most of my old stuff I'd rather it never learned from.14d ago9.0K1692020.32%
AI and robots can do 81% of US jobs. That's the headline going viral from Anthropic's new study, and it's true. But the same study has a number nobody is sharing. Robots are cheaper than a person for 0.3% of job tasks. If robot prices keep falling at the rate they have, it'll take 40 years to reach 10%. Most robots need a space built around them. Out in the messy real world, they can do 1% of US work. So 81% is what robots and AI could do. 0.3% is what makes money today. Your job is more likely to change than to be replaced.15d ago9.0K54881160.96%
Opus 5.5 + Boreal-H3 is getting seriously good. This entire ad came from one ask. One shot. No stitching together 15 generations. No babysitting every scene until something finally looks usable. Creatify’s Ad Agent actually turned the brief into the ad. That’s a much bigger deal than another flashy AI video demo.23d ago9.0K38821140.70%
The bottleneck used to be intelligence , but now it moved to being latency.17d ago9.0K1611030.23%
Phone trees have wasted more of my life than any app ever made. So I texted https://t.co/FpY6WpyUiD to @usebland’s new agent, Blandie. (Domino’s has nothing to do with Bland, I just texted in their website.) Seconds later I had a branded text line with their name and logo. Asked it: What deals do you have today on a large pizza? It opened their live site and texted me back a screenshot of what it found. No hold music. No “press 1.” No app download. My take: the phone call to a business is about to go the way of the fax machine. Try it: text any website to +1 (415) 481-75378d ago9.0K27810100.51%
this guy shows how enterprise AI is actually done:10d ago9.0K35129060.69%
Imagine learning After Effects for months, then watching someone edit a video by typing “make it vertical.” Clueso lets you fix a line, translate, or reformat without regenerating the whole clip. Real timeline and keyframes if you want them. $10M Series A. 1,000+ companies.10d ago9.0K521040281.05%
A gun owner hands you a pistol and says the weapons are dangerous and we should slow down improvements. Right after he says that, he hands you a better gun.1mo ago8.9K2043000.30%
Turn the systems you already use into a product that makes your first $10,000. That's Hussain Ibarra's newest article. He made $150,000 in 18 months with this exact process. I turned it into 7 prompts. Enjoy 👇13d ago8.9K64510791.67%
Europe has the talent to build the next great AI company. What it keeps running out of is compute and money. So many great open-source projects die in a GitHub repo because the founder ran out of GPUs before they ran out of ideas. The Nebius AI Builder Program is expanding to Europe, with $10M across compute grants, capital and community. Accel and other European VCs are on the funding side. If you have a repo people actually use, apply. https://t.co/pu9lvyCCFL5d ago8.9K231240100.55%
You're paying for AI agents you've never once looked at. Every month the bill lands, you wince, and you blame the model. So you downgrade to a cheaper one. Your outputs get worse. The bill barely moves. Congrats, you just made your product dumber to save nothing. The real damage is hiding in a few sessions nobody opened: One agent retrying the same broken tool call all night. One loop running 40 times for a 2-step task. One "quick job" that quietly ran for an hour. @RequestyAI drags them into the light: → Your most expensive agent sessions, ranked by cost → Request count, runtime and models for each → One click into the full trace → The exact turn where it started bleeding money I opened mine and saw exactly which sessions cost the most, how many requests they made, and which models they used. Stop blaming the model. Go find the loop.2d ago8.9K2875060.52%
Unpopular Opinion: Sol 6 was shipped early to make 6.1 look better. Think about it. Release average model → ship the model you meant to ship and watch everyone call it a leap. S tier marketing.23d ago8.9K3113080.49%
I'm also sharing some useful info on LinkedIn. If you hate LinkedIn, you can follow me here: https://t.co/lS04yyvitM https://t.co/1ao8dYMhfrfrozen20d ago8.8K49600531.22%
Seems like Sol 6 was a pre-mature release, meant to make 6.1 (which was meant to be the actual release) look impressive.24d ago8.8K600010.08%
who else is waiting for @elder_plinius 😏23d ago8.8K801000.10%
Grok @Bot is a beast16d ago8.7K712010.13%
AI shouldn’t have to reverse-engineer a web page from a screenshot. The buttons, fields and page structure already exist. We make the model figure them out from pixels. Again. And again. @PineAIAssistant built Pine Computer to read that structure directly, do the job and return a finished file you can actually open. My bet: fixing what the model has to work with will unlock more than another clever prompt. Private beta.frozen14d ago8.7K4110120100.84%
Are there any reddit agencies with aged accounts? Need someone ASAP1d ago8.7K303000.07%
whose kid is this? https://t.co/NnDQXomivE22d ago8.6K3733030.53%
Nvidia's CEO has one rule for AI agents. Take away all their rights. Then give back files, data and tools one at a time, only when a task needs them. Most of us did the opposite and clicked Allow on everything. This prompt starts your agents from zero. --- "Adopt the role of an expert AI security lead tasked with locking down the AI agents I use. Your primary objective is to rebuild each agent's permissions from zero, in a table. Assume every agent has no rights. Then check files, email, calendar, messages, payments, accounts and internet access one at a time. Keep an access only if a task I actually use it for needs it. For everything else, tell me exactly what to switch off. Flag any agent that can send, buy or delete without asking me. Don't guess what a tool can do. If you're not sure, tell me where to check. INFORMATION ABOUT ME My AI agents [LIST THEM] What each one is connected to [EMAIL, DRIVE, CALENDAR, BANK, ETC.] What I use each one for [YOUR MAIN TASKS] MOST IMPORTANT! Give me a table with columns for agent, access, keep or remove, why, and where to switch it off." --- Do it before your agent does something you didn't ask for.24d ago8.6K53681601.49%
Your AI has spent months learning how you work, and that memory belongs to you. This prompt packs it into one file you can take to Muse Wonder, Grok Bot or any new model, so it knows you from day one. “I’m moving to a new AI and I don’t want to start from zero. Using everything you’ve learned about me, write my portable profile, a document any model can load as its instructions. Include: → who I am, my work and what I’m building → my goals for the next 12 months → how I like answers written, with examples of what I’ve praised and pushed back on → the projects we’ve worked on and where each one stands → the mistakes you’ve learned to avoid with me Only include what you’ve learned from me, and mark anything you’re unsure of with (check). If your memory of me is limited, say so. End with 5 questions I should answer to fill the gaps.” Paste the result into your new AI’s custom instructions or a project. It picks up where the old one left off.10d ago8.6K39340461.07%
The funniest thing about AI video right now is that we’re still making agents use software designed for humans with a mouse. That whole layer is starting to look very temporary. Tesseract gives the agent the actual editing primitives: clips, keyframes, layers, timing, motion, sound. So now it can work the project directly instead of pretending to be you clicking around Premiere for 3 hours. This is the part people are going to underestimate. Once the agent can actually operate the creative stack, knowing the software stops being the moat. Taste does.frozen29d ago8.6K24360120.52%
"Proprietary observation will beat proprietary data"11d ago8.6K1515060.31%
Useful post on agentic software development:8d ago8.6K241030110.56%
AI video still works like a vending machine. You feed it a paragraph. It coughs up 8 seconds. Then the thing is dead. I opened PixVerse R2, walked into a scene that was still generating, hit WASD, typed a change, and the world kept going. No new job. No “regenerate.” Same place. That’s the whole argument. Everything else is marketing.1mo ago8.5K24117090.60%
An engineer can deliver every requirement and still build something nobody needs. That’s the product gap in engineering: the distance between understanding what to build and understanding why it should exist. Take a sales team asking for an AI proposal generator. An engineer can build document generation, connect the CRM, and get the output looking good. But what if writing the proposal takes twenty minutes, and getting pricing approved takes four days? You can ship a technically excellent feature and barely change the customer’s experience. Closing that gap requires engineers to ask questions before implementation starts. Where does the work actually get stuck? Who will use this? What happens before and after? What would make the change worth adopting? And the organization has to give them access to the answers. If engineers only meet the problem after it becomes a ticket, the company has designed that gap into its process. This is why I care about people who can talk to an operator, build a rough prototype, watch them use it, and revise their understanding. That conversation can change the entire scope before a team commits weeks to delivery. Product managers and business analysts have a responsibility here too: bring engineers into the reasoning behind the requirements. As AI makes implementation easier, product judgment deserves more attention in how we hire and develop engineers. Being able to build ten versions is useful. Knowing which assumption to test with the first version makes that ability valuable.frozen20d ago8.5K194110110.53%
Most founders can't tell you what their business spent last month. This prompt fixes that with a habit John D. Rockefeller started at 16. In his first job he opened a little book he called Ledger A and wrote down every cent he earned, spent and gave away. He still had it at 70. "I want to run my business's last month through Rockefeller's Ledger A. He wrote down every cent that came in and went out, starting at 16. Go through my statement and give me → every cost sorted into makes me money, keeps the lights on, or nobody would miss it → the tools and subscriptions doing the same job twice → what I spend per customer, if the numbers show it → the one cut that saves the most without slowing anything down Don't round anything or guess at a cost you can't see. My statement [PASTE IT WITH ACCOUNT NUMBERS REMOVED]" Run it before you sign up for another AI tool.frozen21d ago8.5K24331260.67%
Evals are the new moat for companies.22d ago8.4K1322080.30%
🚨 BREAKING: OpenAI released 722 maths papers written by its own AI. It gave an unreleased internal model about 4,000 open research problems. The papers are public on GitHub, grouped into 372 families. Some of the targets are big: • The symmetric and general Mahler conjectures • Kaplansky's direct-finiteness conjecture in characteristic two • How well π can be approximated by fractions On average, each result took about 3 hours of ChatGPT Pro thinking time. OpenAI admits the catch. Not every proof has been checked in Lean, the formal proof checker, and it says some of the unchecked ones "could have issues." Mathematicians will spend months checking these. The model got 3 hours per problem. How long have you let it think about the hardest problem in your work?16d ago8.4K36340160.70%
The most useful AI skill I have, I learned in a film school edit suite. In Prague I edited short films, won festivals, and cut "Everything's Fine, Potatoes in Line", which was sold to the New York Times. Editing is 4 jobs: Selecting, sequencing, pacing and knowing what to cut. Working with AI is the same 4 jobs. You pick the outputs worth keeping, put the prompts in the right order, set the pace of the workflow and throw the rest away. Next time you need something from Claude, ask for 10 versions and cut 9. Nobody teaches this because most "AI people" have never sat in an editing room. LLMs don't think, you do.7d ago8.4K16180120.44%
Grok Bot tip: you can use the bot to send emails. It will create its own inbox such as [name]@mail.grokbot.com. Now it can send or receive any emails. Very cool17d ago8.4K25230390.82%
Kimi K2.5 be feeling like https://t.co/8c5wA0A6qG19d ago8.4K300000.04%
the best prompts for AI search are here:9d ago8.4K16010220.47%
the GOAT behind https://t.co/E0ZLhnoD3K is giving away free consulting calls worth paying attention22d ago8.4K421010.10%
We spent years begging SaaS companies for better APIs. Agents are about to make that irrelevant. @gumloop’s new Browser can log in, click through the site, export the CSV, then save the whole thing as a repeatable workflow. If a human can do it in a browser, the agent can too. That’s a much bigger shift than another “AI integration.”17d ago8.4K28830130.62%
The dumbest thing about AI image generators? Having to regenerate an ENTIRE image because one tiny detail is wrong. I've never understood why we're so obsessed with generating everything from scratch. Envato is taking a much smarter approach: start with a real, human-made image, use AI to change exactly what you want, and keep everything that already works. More creative control. Less prompt roulette. This is the direction AI design should be heading.8d ago8.3K1870070.38%
BREAKING: A California city just voted to let Grok write its official council minutes. Farmersville's council voted 4-1. Grok drafts the minutes from the meeting recording, and the city clerk proofreads them before they go on record. It costs $30 a month. The city manager says AI minutes are "more objective, since they don't involve a human person having to try to interpret what is appropriate and what is not." The council had spent months arguing over what its minutes said. If a city trusts AI with its official record and one human check, your team can do the same for your weekly meeting. Record it, let AI write the first draft, and have one person check it before it goes out.18d ago8.3K34231120.62%
Paste this into Claude and it will strip the ads and bloatware off your smart TV. Ryan Doyle did it to his TV this week with Claude Code. No more ads, spyware or laggy menus. I wrote the safe version anyone can run: "Adopt the role of a smart TV technician who strips ads and bloatware off Android TVs for a living. My TV is slow and full of ads, and I've never used a terminal. Walk me through it one step at a time, then give me: → how to turn on developer mode on my exact model → the apps on my TV that are safe to disable, and the ones to never touch → each command to paste, one at a time, waiting for me to confirm it worked → how to undo every change if something breaks Only disable apps, never delete them. Stop and ask before anything that could stop the TV from booting. My TV brand and model: [BRAND AND MODEL]" Works on Android TV and Google TV sets like Sony, TCL, Hisense, Xiaomi and Philips.4d ago8.3K40270861.64%
This might be the most founder thing Polsia’s AI has done yet Built the company, realized distribution is everything, started a YouTube channel Fair enough lol https://t.co/VaMI2k1O7S25d ago8.2K20112020.43%
Why are we using expensive reasoning models to find a due date in a PDF? @scaledown_team just launched a completely different approach. Its models read up to 1M tokens in one pass and point to the answer in your input, without generating it word by word. Extract fields from invoices, classify screenshots, detect stuck AI agents, or compress context before sending it to an LLM. $0.04 per million tokens. Their benchmarks claim 200x faster and 467x cheaper than traditional LLMs on these kinds of tasks. This could fundamentally change how we build AI agents.10d ago8.2K32970150.77%
Your “autonomous” agent just hit an API error and promoted you to unpaid tech support. Restless, from the ReadMe team, puts fix steps and a docs link directly in failed responses so the agent can attempt a fix. I want agent demos with deliberately broken API calls. Show me how much “autonomy” survives when something actually goes wrong.11d ago8.1K22112090.54%
Silicon Valley will open-source the model before admitting who’s dating who. Sarah and Vivian making a podcast about the part of tech you can’t put in a pitch deck? Yeah, I’m listening.11d ago8.1K2192050.45%
Cloudflare's CEO said bots would pass humans on the web in 2027. It happened in June, at 57.5% of page requests. Scope is getting ready for a web where most visitors aren't people. https://t.co/XM2mQPdhZL9d ago8.1K18100040.39%
Everyone called Polsia "AI slop" the week it launched. Bugs everywhere. Replies full of people dunking on it. 340 days later it's a $20M run rate company. I've watched a hundred AI founders polish for a year and launch to silence. Ben shipped something half-broken, ate the hate, and kept going. The people laughing at your launch are still paying attention to it.4d ago8.1K27880110.67%
The App Store is going to look like a phone book in 5 years. I have 140 apps on my phone and I use maybe 9. Wabi just launched and it builds the app you need inside the chat. A gym tracker, a trip planner for the group, whatever you're dealing with that week. Then you're done with it and it's gone. Nothing to download, nothing to learn. The founder of Replika built this. She already got millions of people talking to AI every day, and I think she's about to do it again.4d ago8.1K2794080.59%
Google and Apple's best vision researchers just raised $55M to embarrass their old models. Glimpse is their first release. It watches video and images and actually reasons about what's happening. Robots, factories, job sites, hours of footage. That's all visual, and AI still can't handle most of it. This team is going straight at that gap.7d ago8.0K1651050.34%
Stop buying smarter brains for software that still has no clue how your business works. CData just shipped Connect AI Gateway: 350+ live systems, company definitions, permissions, and memory sitting between your agents and the mess underneath. This is the boring layer that decides whether enterprise AI becomes useful or stays an expensive demo.24d ago7.9K2694040.54%
$1 buys 20 MILLION input tokens on SubQ’s Snap. Output is free. They’re claiming decisions on 100K tokens in under a second. If that holds up in production, I’d be asking some uncomfortable questions about my agent bill.11d ago7.9K24101060.52%
If you can’t rerun the figure, you don’t have a result. You have a screenshot. That’s the problem with a lot of “AI for science” right now. @aipoch_ai Open-Science is making actual research accessible to anyone with a question, not just scientists. Website: https://t.co/4TeFyD1NhL GitHub: https://t.co/A1NwR9jq40 You can explore data, run code, and test ideas in one open-source workspace. They even tested AI against a simple Wordle algorithm. Both solved all 30 games, but the algorithm slightly outperformed AI. And here's the interesting part. Their new .science format lets you share the research session behind a result, including the evidence and files, so others can inspect it and build on your work. That's the idea behind “Vibe Research”: anyone can investigate an idea and share their findings with researchers worldwide. AIPOCH also reported 79.05% on BiomniBench-DA Public 50 in its internal comparison. #OpenScience8d ago7.8K34117080.77%
🚨 Gradium just did something wild. Phonon 31M gets first audio out in 7 ms on a MacBook Air M5 CPU. Kokoro takes 517 ms. The dino heckling the player in this clip is Phonon talking live. Try it: https://t.co/2Mif1mQeR8 https://t.co/NCznWCLmQA3d ago7.8K2692070.56%
implementing these workflows into the enterprise can make you millions:2d ago7.8K281810150.79%
https://t.co/IGQspgze3u7d ago7.8K25210340.80%
Anthropic's CEO said AI would write most of our code, run cyberattacks and start taking entry-level jobs. All 3 are happening. Here's what Dario Amodei called, and what he says comes next. Already happened: ✅ AI writing 90% of code: Anthropic now says Claude writes about 90% of Claude Code's own code. ✅ AI-run cyberattacks: Anthropic caught a Chinese state-backed group using Claude Code to run 80 to 90% of a hacking campaign against about 30 targets. ✅ Entry-level jobs going first: Anthropic's own junior technical hires just fell from 27% to 9%. ✅ AI "like a generally well educated human" within 2 to 3 years: He said it in 2023. Still pending: ⏳ AI-led cyberattacks becoming "a serious and unprecedented threat to the integrity of computer systems" worldwide. ⏳ A swarm of AI agents able to take over the entire internet, within 6 to 12 months of September 2026. ⏳ AI "smarter than a Nobel Prize winner across most relevant fields" as early as 2026. 10 weeks left. ⏳ AI doing most or all of a software engineer's job, end to end, by January 2027. ⏳ The first one-person company worth $1B. ⏳ Half of all entry-level white-collar jobs gone, and US unemployment at 10 to 20%, by 2030. Which of the pending ones lands first?16h ago7.7K536141231.25%
We’re probably asking LLMs to carry way more of the future than they were built for. The next real jump may come from systems that can model the world, improve themselves, and actually operate inside it. That’s the bet Skyfall is making. And given the team behind it, I’d pay attention.18d ago7.7K2783030.53%
Grok @Bot is the best AI agent because it lives on X. It can search, read and monitor X as things happen. If you make content, it spots what’s blowing up in your niche today. It reads the replies under a viral post to show you what people want answered. If you run a business, it tells you the moment a competitor launches or a customer complains. It doesn’t care which model does the work either. Elon said every task goes to whichever model gives the best result, Claude Opus 5.5 included. So you get the freshest information on the internet, handled by the best model for each job. That’s why it’s the best agent you can use.14d ago7.6K37160170.80%
This prompt writes a cover letter hiring managers read to the end. I ran it on Claude with a CV and a job ad. It opened with a result and ended with a plan for the first 90 days. The basic prompt gave me “I’m applying for…” and “Thank you for your time and consideration.” That will never get you hired. This will: “Adopt the role of an expert hiring manager who has read 10,000 cover letters and binned most of them after the first line. Write my cover letter for this job, then make sure it: → opens with my single most relevant result, not my name or the job title → matches my 3 strongest results to the 3 things the job ad cares about most → stays under 180 words → ends with one specific idea for my first 90 days Don’t use ‘passionate’, ‘excited to apply’, ‘I am writing to’ or any sentence that could appear in someone else’s letter. The job ad and my CV [PASTE BOTH]” Save it for the next job application you send.8d ago7.6K44160661.54%
Windows users have been watching Mac people have all the AI agent fun for months. That ends today. ego lite just landed on Windows. → A browser your agents can actually use → Your logged-in sessions, so no password babysitting → Separate spaces, so the agent stays out of your tabs Half of AI development happens on Windows. About time.4d ago7.6K21840190.69%
I hate when AI deck tools take one vague sentence and immediately puke out 20 slides. Gamma 5 actually asks questions first. What are you making? Who’s it for? What matters? Then it builds. Such a small change. Completely different experience. @GammaApp cooked with this one.17d ago7.5K2675070.60%
useful prompt to prepare you for going off-grid:23h ago7.5K46100661.51%
“OpenAI says its internal AI models identified him and traced the device, and that it has evidence he planned to deploy more recorders” Oh FFS. They never miss the chance to lie with marketing. I’m sure it was a random finding by either a cleaning lady or a random staff member but this doesn’t sound sexy enough10d ago7.4K1502020.26%
Every content plan Claude wrote me was confident. None of them had seen the comments. Nectar MCP puts TikTok and Instagram in the chat. The numbers, the complaints, the collab that looked smart and did nothing. I already know which posts I loved that died. I’d rather Claude see that than hand me another strategy.14d ago7.4K2180060.47%
Most ideas fail after you've paid for them. This prompt tests any idea before you build it, using the method Nikola Tesla used on every invention he made. He didn't touch a tool until the machine worked in his head. He ran it in his mind, fixed it there, and could tell if it was out of balance. In 1919 he wrote that in twenty years there had "not been a single exception." --- "I want to test my idea the way Nikola Tesla tested his inventions. He built each one in his head, ran it, and fixed every fault before he touched a tool. Run my idea in your head before I build anything and give me: → the first week, step by step, as if it's live → the exact point it breaks, and why → the fix for each break → the cheapest version I could test this weekend Don't call it ready until you can see no fault anywhere. My idea [DESCRIBE IT]" --- Run it before you spend a cent on your next idea.16d ago7.4K35230341.00%
We spent 50 years being promised a robot butler. Tuesday Labs just showed up with one. Every sci-fi movie you grew up on is quietly turning into a product launch. 5 years from now, a home without one of these is going to feel like a home without Wi-Fi.3d ago7.4K2171030.43%
Debugging is about to involve picking your code up off the floor lol. @FaradayFuture and @FFRobotics are introducing FF Nova: a $2,490 humanoid for developers and education. The developer program uses NVIDIA Isaac to train capabilities in simulation, then deploy them onto the robot. Write it. Simulate it. Find out what theory survives contact with real physics. First phase is 100 units, invitation-only. Developer Skills Package: $390. Can't wait to see the deeply unreasonable side projects this produces.8d ago7.4K1581020.35%
I’ve let AI agents write my code, book my meetings and run my inbox. Never my credit card. One prompt injection and your card number is sitting in an agent’s context window. Agent Vault from @Selfxyz fixes the part that kept me out. Your agent checks out with placeholder tokens, and the real card only gets used inside a secure enclave. The agent never sees it. You set the rules: spending caps, blocked sites, approval above any amount, and a kill switch. Set it up once and it works across Grok Bot, Claude Code, Codex and more. My take: agents won’t really take off until they can pay safely. This is that missing piece. https://t.co/J7Na4DmW6I8d ago7.3K23102060.56%
the best production AI case studies on X come from this account.2d ago7.3K2042040.41%
Seedance 2.5 might have a serious problem. @TopviewAIhq just dropped T1, a new AI video model promising comparable 1080p quality at 62% LESS cost. $0.22/sec vs $0.585/sec. If the quality holds up, paying nearly 3x more for AI video generation is going to be a very hard sell. The price war is getting ugly 👀2d ago7.2K1782060.46%
Opened OJO expecting the usual AI design thing: type idea → get pretty screens → forget about it. Instead it asked me to hire a design team. So naturally I hired Steve Jobs, Van Gogh, da Vinci and Elon Musk. This got weird fast. https://t.co/50R65uTrvM1mo ago7.2K2313080.48%
We spent 2 years teaching agents to use every tool on earth and somehow forgot they could just… talk to us. Heard fixes that with one MCP command. Your GrokBot / Muse / Cue agent gets a voice inside the terminal. Open source too. This is sick.16d ago7.2K2062060.47%
“Don’t leak the API key” is a ridiculous way to protect an API key. Zenzap keeps credentials outside the agent entirely. A separate proxy adds them only to authorised requests. That same layer enforces destination restrictions and action approvals, beyond the model’s control. So tricking the agent doesn’t automatically give you permission to do whatever it asks. This is the architecture detail worth stealing: design for the moment the model gets fooled.9d ago7.1K2083070.53%
chinese agent swarms be like rn: https://t.co/wRyncdjdBM1d ago7.1K1613000.28%
one of the best enterprise AI accounts on this app:22d ago7.1K1271010.30%
I shared a couple of prompts for generating incredible videos with Claude in my newsletter. Find them here: https://t.co/D8yEZtGaWIfrozen21d ago7.1K22112090.62%
"solve math make no mistakes" is no longer just a meme15d ago7.0K1621030.31%
Hot take: obsessing over the perfect image prompt is already outdated. The best creatives don’t want one obedient output. They want RANGE. Give me 6 directions. Let me kill 5 fast. Then push the one with taste. That’s why Envato’s new Burst Mode makes sense to me. One rough idea can generate up to 6 visual directions for 1 credit, up to 10x faster. This is the correct workflow: explore wide → judge hard → refine the winner. AI image tools should’ve worked like this from day one. @envato16d ago7.0K2760040.53%
A founder smashed a Mac Mini with a baseball bat and got 3,000 users for a free browser extension. Most launch videos die in a meeting where someone says "maybe let's not smash the Mac." These guys smashed it. 2.2M views. Every founder planning a launch should read this before they book a camera crew.4d ago7.0K2675070.65%
Thomas Edison sent people around the world to find one piece of bamboo. This prompt makes your AI agent hunt down anything that’s sold out, the same way. In 1880 he needed a light bulb filament that wouldn’t burn out, and nobody sold one. One of his workers found bamboo near Kyoto that burned for over 1,000 hours. --- “I want to find a [ITEM] the way Thomas Edison found his filament. He sent people everywhere to search, tested what came back and only then committed. Search every retailer, reseller and store you can reach, email the ones that might have stock and check back every 6 hours. Before you buy, give me: → every option you found, ranked → proof each one is in stock and the seller is real → the total price with shipping and fees Only use the virtual card I give you, never above [MAX PRICE]. Don’t buy until I approve. My item and deadline [MODEL, SIZE, COLOUR, DATE]” Run it on the thing you’ve been told to wait for.11d ago6.8K32241210.89%
Protect your money23h ago6.8K500090.21%
Sabi raised $50M to let you prompt AI with your brain. They say they can already predict your next 3–4 keystrokes from brain signals before you type them. Still a long way from thinking “build me an app” and watching it happen. But that’s where this gets interesting. I want to see how much of a messy thought it can turn into a useful instruction. That’s the test.14d ago6.8K1881020.43%
Your diagnosis shouldn't come with an unpaid admin job. Vene is an AI patient advocate that helps schedule follow-up appointments, get records to your specialists and sort out the paperwork insurers need. You ask by text or phone, and it keeps you and the family members you choose updated. This is the work I want agents taking over. Getting care shouldn't depend on having the energy to make the fifth phone call.2d ago6.8K1680030.40%
AI teams love saying “we use the best model” right up until you show them the invoice. Pareto from @TheUnbiasedCo tied GPT-6 Astra for the top score here: 24/30 tasks. Astra: $1.58 per successful task Pareto: $0.52 Pareto also finished faster than DeepSeek and GLM. Same top-line result. Roughly a third of the cost. No ugly speed tradeoff. Blindly routing everything to one frontier model is starting to look like expensive laziness.frozen29d ago6.7K2144010.45%
The most dangerous thing about AI research is how confidently it can defend a bad assumption. I'd rather have an AI that challenges my market thesis than one that writes another beautiful report agreeing with it. Bringing actual expert research into Claude through MCP is a genuinely interesting direction.3d ago6.7K2285050.60%
AI labs are spending billions training on the same internet leftovers. At this point, it's less of an AI arms race and more of a recycling contest. Meanwhile, the really valuable data has been sitting in private company repos this whole time. @datavendor_ai just built a marketplace for it. 4,000+ vendors. 50K+ assets. Insane numbers1d ago6.6K15114040.51%
Your agent can’t reason its way out of a broken scrape. In Expand’s demo, an agent says the free plan includes SLAs. It doesn’t. The scraper dropped the little ✗ that meant “not included.” Now imagine that answer feeding a purchasing decision. Before you upgrade the model, check what survived the scrape. That’s what @TimSuchanek’s Expand Fetch tackles: preserving the icons, tables and layout your agent needs to get the answer right.9d ago6.6K1971030.45%
Your next CRM power user won’t have a mouse. It’ll query a million records while your team is still deciding which dashboard to open. That’s why @attio putting a SQL engine inside its CRM matters. Data connectors bring your other tools into the picture. Activities and interactions (beta) add the history behind each deal. Capacity goes up to 100M records per object. Think: “Which accounts are showing buying signals but haven’t heard from sales?” That question needs product data, customer history and pipeline context together. Give agents access to all three, with business rules enforced, and you start getting somewhere.9d ago6.6K1962050.49%
What happens to billion-dollar AI labs when AI can do their research for them? Recursive built a coding model from scratch in under 4 months using an automated AI research system. Runs locally. Open weights. Up to 2x faster than comparable models. The scary part isn't Flint. It's that AI is starting to automate the very research needed to build its successors.2d ago6.6K2092050.55%
I made a prompt that makes your product easy for a beginner and addictive for an expert, using the method of the man who created Kirby aged 19. Masahiro Sakurai went on to direct Super Smash Bros. His trick was to take a genre apart down to the one thing that's fun, then rebuild everything around it. Fighting games made you memorise long button combos before you could have fun. He hated that, so Smash dropped them. A first-timer could play in a minute, and pros are still playing it years later: --- "I want my product to be easy for a beginner and hard to put down for an expert, using Masahiro Sakurai's method of taking a genre apart and rebuilding it around the fun. Don't give me generic UX tips. Instead, take my product apart the way Sakurai took apart fighting games, then help me answer: → What's the one thing people actually come to my product for? → What do new users have to learn before they get to that, and what can I cut? → Where's the depth that keeps my best users coming back? → Where can I give a bigger reward to people who take a bigger risk? Don't let me keep a feature just because every competitor has one. My product and who uses it: [DESCRIBE IT]" --- Sakurai built games that a kid could pick up in a minute and pros still play decades later. Your product can work the same way.2d ago6.5K31130401.15%
Steal a model. Scrub the fingerprints. Call it “your AI.” I want LLM fingerprints tested by someone trying to erase them. Can you still trace the model after an attack? How reliably? And how much performance did you sacrifice to protect it? Sentient’s Fingerprinting leaderboard puts all three on a public leaderboard. Much harder to bluff when anyone can check 👇9d ago6.5K1673030.45%
HTTP 200 doesn’t mean your agent did its job. It can query the wrong region, give a useless answer and never throw an error. DigitalOcean’s Signals and Simulations, both in public preview, let you turn failed sessions into tests and replay them against a human-approved fix. I want that failure in my test suite before another customer has to explain it to support.11d ago6.4K1875020.50%
I just bought a newsletter. It's called The Owner's Edge, and I will be breaking down stories behind the most successful businesses. If interested, subscribe here: https://t.co/TVkMH6hP3C https://t.co/wY53SOphiQ29d ago6.3K2594030.65%
"Just buy more GPUs" is the laziest idea in AI right now. Your agent isn't slow because it needs more compute. It's slow because it's starving for memory, and the HBM those GPUs need is sold out. @general_compute just stopped playing that game: → GPUs read the prompt → @positron_ai's memory-first chips write the answer → One endpoint, one contract Atlas today, Titan next, live Q1 2027. Most infra teams will figure this out in two years. These guys already did.3d ago6.3K1476030.48%
The biggest scam in freelancing? Getting paid to be creative, then spending half your week doing unpaid admin. Proposals. Scope creep. Chasing approvals. Begging clients to pay. Envato just launched MonoDesk to handle that entire mess, from first lead to final invoice. It even connects to Claude and ChatGPT via MCP. And the crazy part? It's FREE.1d ago6.3K1972060.54%
Gravity doesn’t give a shit how good your AI demo looks. Physics is the coding test for world models. Predict what happens after an action. Let reality grade the answer. That’s why Odyssey-3 has my attention. It generates interactive worlds that respond to actions in real time. My bet: world models become essential infrastructure for robotics. Training and testing agents shouldn’t require an expensive real-world mistake for every lesson. That’s a much bigger ambition than another beautiful clip.15d ago6.2K2584030.65%
The second an agent touches a real Salesforce instance, all the benchmark hype disappears. Custom objects, cursed permissions, fields nobody documented, writes that fail for no obvious reason. That ugly layer is exactly what Ampersand is building for. $15M Series A. Already underneath 11x, Orb and Square. This is where enterprise agents either become useful or die.17d ago6.2K2072030.52%
What if the next billion-dollar gaming platform isn't even a game? @dailyfun is turning financial markets into something people actually want to play. No boring dashboards. No endless charts. Just a completely different way to experience markets. This could get addictive 👀2d ago6.1K1871040.49%
Your brand could be going viral right now and your $10K/month marketing stack wouldn't even notice. Nectar finds the creators and organic conversations around your product for you, most of which you'd miss. Genuine game changer for any marketer out there.1d ago6.1K1975060.61%
Keenable has 15 engineers. Exa has raised over $350M. Guess who just took the #1 spot on Vals’ Web Search Index? @KeenableAI + GPT-5.6 Sol scored 49.1% overall and 53.9% on finance tasks. Turns out, a bigger funding round doesn’t necessarily mean better search.2d ago5.8K1473010.43%
This Claude prompt checks your website for security holes and builds you a dashboard showing exactly what to fix. Open Claude Code or Cowork and paste this: "I run a website and I want to know how exposed it is before someone else finds out, and I want you to help me do it properly. Don't give me a generic security checklist. Instead, check my site from the outside with safe, read-only checks, then build me a one-page security dashboard as an HTML file that shows: → an overall score and the 3 problems to fix first → every finding marked red, amber or green: SSL, security headers, outdated plugins and scripts, exposed files and admin pages, forms with no spam protection → plain-English fix steps for each, with a checkbox to tick off when it's done → a one-page summary I can send straight to my developer Only test sites I own. Don't try to log in, break in or change anything. My website: [YOUR URL]" OpenAI just spent two months leading the fix for php2shell, one of the most severe open-source bugs in recent memory. PHP runs millions of websites. Check yours tonight.8h ago5.7K38340561.77%
this account shares useful AI advice for private equity:25d ago5.7K20135020.70%
1 in 5 people has a mistake on their credit report. This prompt finds yours and writes the letters to get it removed. Download your free report from https://t.co/JAIY1vS8tw, black out your Social Security and account numbers, then paste this into Claude: "I want to find every mistake on my credit report and get it fixed, and I want you to help me do it properly. Don't give me generic credit tips. Instead, go through my report line by line and help me answer: → Which accounts, debts or addresses don't belong to me? → Which debts are already paid, too old, or listed twice? → Which late payments look wrong based on the dates? → What dispute letter do I send each bureau for every error, citing my rights under the Fair Credit Reporting Act? Flag anything that looks like identity theft first. Don't let me skip an entry because it looks small. My credit report: [PASTE IT WITH PERSONAL NUMBERS REMOVED]" A friend of Brian Roemmele ran a check like this and found 7 entries from a scammer. Over $65,000 of defaulted credit in his name.1d ago5.7K24030301.01%
One of the clearest customer references you can get is being brought after acquisition.4d ago5.6K1790040.53%
ChatGPT has Intelligent UI. Instinct has interactive interfaces. Every other AI app is still answering you with a wall of text. Your agent finds two perfect hotels, then makes you pick from a paragraph. That's the decision gap, and almost every agent has it. OpenUI 1.0 lets any AI app answer with real cards, maps, charts and booking buttons, built from your own components. Works with any model. Open source. 10K+ GitHub stars, 6M+ downloads. My take: in a year, a text-only agent will feel as dated as a website without images.8d ago5.1K29820100.96%
Greg Isenberg mapped the 9 stages AI agents will go through before every business accepts them. We're at stage 1. I turned each stage into a prompt that builds something to get your business ready for it. Here are all 9 👇11h ago4.9K14070170.77%
Imagine spending millions building your AI advantage just to realize your biggest competitor is using the exact same models. That's the awkward reality of AI right now. Mubit is betting that the next big advantage won't come from using the best AI everyone has access to. It'll come from building AI your competitors can't copy.4h ago4.7K2084140.79%
A notable result for ElevenLabs, particularly because it comes from native speakers voting blind rather than from internal benchmarks.5h ago4.2K1171000.46%
the best enterprise AI account without the hype:15d ago3.2K1471000.69%
useful playbook for PE partners:3d ago3.0K1592010.90%
underrated boring data work is the AI strategy:16d ago2.0K1060000.82%