科技热门推文榜——2026年9月19日

今日科技要闻：研究人员声称，两项漏洞可导致 ChatGPT 和 Codex 账户被接管，并可能暴露与之关联的企业工作服务，令 AI 安全成为讨论焦点。新近发布的产品包括隐私保护工具 AgentCloak、法律平台 Astra for Law，以及 AgentSky 推出的浏览器智能体市场。工程领域的讨论则聚焦于超紧凑型本地模型、经审计的 AI 基准测试，以及利用热量在物理隔离计算机之间进行通信的技术。与此同时，创作者正探索 AI 辅助视频制作和长篇叙事；经过精选的工具合集与自动化课程也在持续推动 AI 的实际应用。

1. ShakeelHashim (Group Score: 561.7 | Individual: 58.0)

Cluster: 20 tweets | Engagement: 2184 (Avg: 100) | Type: Tech

RT @S1r1u5_: On July 25, we hacked OpenAI.

Two bugs let us take over ChatGPT/Codex accounts of OpenAI employees (+some unaffiliated users) and reach connected services: Outlook, Slack, GitHub, etc.

We proved it with a PR in OpenAI’s internal codebase . It took us <72h. 🧵 https://t.co/gVsmQZwSc8

See 19 related tweets

@WesRoth: whoa...

it's getting real\n\nQT @S1r1u5_: On July 25, we hacked OpenAI.

Two bugs let us take over ...

@cryptopunk7213: we need to wake up to the very real threat of ai cyber attacks.

they used opus 4.8 and 5 to exploi...

@JeffLadish: If they got full access to the monorepo, that means they could have downloaded OpenAI’s entire code ...
@SebJohnsonUK: The very first team backed by @ProjectEurope_ successfully hacked @OpenAI.

Here's how:

team bega...

@cramforce: When I might have sounded alarmist over the last few weeks, it was because I was aware @S1r1u5_'s ex...

2. LearnWithSubhan (Group Score: 419.5 | Individual: 40.0)

Cluster: 14 tweets | Engagement: 212 (Avg: 76) | Type: Tech

RT @zakiraicoder: 50 legendary websites that feel like the internet’s hidden toolbox 🧰

https://t.co/IF0e93Vsik — Free research papers
https://t.co/hJt2YM0nzp — Borrow books online
https://t.co/VEPPx9b4Jw — Free academic journals
https://t.co/9Y8pcXCzKu — 70K+ free books
https://t.co/wYZrW6O9C0 — Free textbooks
https://t.co/Bobq0w1hyR — Free online courses
https://t.co/evZc7gnIRO — Solve complex problems
https://t.co/9Q3muzXKRO — AI research assistant
https://t.co/UNJRIStN25 — Research-backed AI answers
https://t.co/yS5IABTRNh — Find research connections
https://t.co/6ZduOfAU1p — Academic search engine
https://t.co/e0pHZRAc7r — Understand research papers easily
https://t.co/siSTGmfU8y — Free Photoshop alternative
https://t.co/HpcKqPbpyZ — Compress & optimize images
https://t.co/viSOVfImJd — Remove image backgrounds
https://t.co/YzQ27OlQ2m — Remove objects from photos
https://t.co/eebiOp0kCd — Remove video backgrounds
https://t.co/spsEfKcCzi — Beautiful code images
https://t.co/r8rxbML0O9 — Share elegant code screenshots
https://t.co/a4gnOX6DPv — AI search engine for developers
https://t.co/DWO81MQyqe — Test and debug regex
https://t.co/3b1OAJQlrn — Format & clean code
https://t.co/DhBWOvxNYH — Format & validate JSON
https://t.co/9v0uL8DCzG — Understand terminal commands
https://t.co/xXhy4Iacmd — Stunning product mockups
https://t.co/C5jBu0nZpA — Powerful online mockup generator
https://t.co/BdlNvpK4nQ — Find alternatives to any app
https://t.co/HUfTXu5MjB — Check data breaches
https://t.co/J5nf7cVzgN — Scan files & URLs for malware
https://t.co/JwqMSHZ311 — Self-destructing text notes
https://t.co/zSKCYd0ViU — Disposable temporary email
https://t.co/tHSNS9bUnh — 10-minute temporary email
https://t.co/ekKTp2Uksd — Temporary & secure file sharing
https://t.co/7JjneNRlN5 — Explore global live radio
https://t.co/RDJz7xtPwp — Discover similar music & artists
https://t.co/dLZYENDpkU — Find songs from movies & shows
https://t.co/pBcnacpBD8 — Focus music for coding
https://t.co/sMHB2bOHEq — Custom background sounds
https://t.co/UHoFN2OLIw — Ambient café sounds for productivity
https://t.co/XyhBiTbzTH — Find where to stream movies
https://t.co/KDjOuBVmft — The internet’s digital library
https://t.co/5HALmVhiN0 — Create a personal copy of any webpage
https://t.co/1LgssfQ1z7 — Find related websites
https://t.co/oGs5Rw3OnH — AI YouTube video summaries
https://t.co/dRO90EfwYJ — Ultimate smart bookmark manager
https://t.co/MSjW4Fe1JF — Check real-time service outages
https://t.co/pdSYz9s99p — Smart reverse image search
https://t.co/mOEq7HaMX8 — Quick internet speed test
https://t.co/U9L4qe5fIY — All-in-one PDF tools
https://t.co/q35tv4QVSL — Merge, split & convert PDFs

Save this. You’ll definitely need some of these later! 🔖

Follow @zakiraicoder for more useful websites, AI tools & tech resources.

See 13 related tweets

@LearnWithSubhan: RT @justinbrave21: 50 websites that feel like the internet’s hidden toolbox 🧰

https://t.co/w5vhK...

@monicaa_AI: RT @NextGenAi5: 50 websites that feel like the internet’s hidden toolbox 🧰

https://t.co/TFAuWUxl...

@mr_nirajkumar07: RT @Tech_Tafsin: 50 Web Sites Google Doesn't Want You to Know About

1.) 'https://t.co/eDb3nGQjDl' —...

@LearnWithSubhan: RT @marcus_techAi: 50 websites that feel like the internet’s hidden toolbox 🧰

https://t.co/HABav...

@mr_nirajkumar07: RT @monicaa_AI: 50 legendary websites that feel like the internet’s hidden toolbox 🧰

https://t.c...

3. Rubab59f (Group Score: 409.5 | Individual: 32.2)

Cluster: 23 tweets | Engagement: 178 (Avg: 83) | Type: Tech

RT @ethan_walker_ai: Your camera roll might already have everything you need to create better content.

GPT-6 Astra + DaVinci Resolve turns existing photos and clips into structured videos with editing, captions, and voiceover.

Sabrina’s guide includes a practical walkthrough and a prompt you can adapt to your own business. Worth saving. 🚀

See 22 related tweets

@ecoleai: RT @JacobyoungAI: Most businesses already have more content than they think : it’s just sitting in t...
@mr_nirajkumar07: RT @sonalshukla3377: A lot of businesses already have enough photos and clips sitting on their phone...
@robertsmith_ai: RT @Lupin_Ai_Coder: If your business photos and clips are sitting on your phone, this is worth savin...
@robertsmith_ai: RT @Ayzacoder: A folder of coffee shop photos and clips became 3 different videos with GPT-6 Astra +...
@anjum_ai: RT @the_futureAI: AI video editing just got a serious upgrade.

GPT-6 Astra + DaVinci Resolve can tu...

4. Tech_by_Shweta (Group Score: 398.2 | Individual: 59.5)

Cluster: 11 tweets | Engagement: 718 (Avg: 28) | Type: Tech

RT @peteryared: Introducing AgentCloak: Use any AI without sharing your real data.

Chinese AI services, ChatGPT, Claude, doesn't matter.

You probably try to hide details before asking: different names, fake numbers, no address.

But then the answer's useless because the AI is missing actual context.

AgentCloak runs in your browser.

It swaps your sensitive info for realistic fakes before sending anything, then swaps your real info back into the response.

You get what you need.

The AI gets nothing about you.

Works entirely in-browser.

Already trusted by some of the biggest companies in the world.

Now free to use.

https://t.co/hioia521QQ

See 10 related tweets

@shiri_shh: sir....we can't collect users' personal data ANYMORE

they just launched a tool that swaps everythin...

@aakashgupta: AgentCloak keeps the real you out of every prompt you send.

The workaround most people run today is...

@alex_prompter: Chatgpt, claude, gemini, grok and deepseek are all open on this machine.

I’ve pasted real numbers i...

@omarsar0: AgentCloak is a free browser extension for handling personal details in AI prompts.

It replaces det...

@DataChaz: FAKING YOUR DETAILS MANUALLY RUINS THE AI OUTPUT

Meet AgentCloak

AgentCloak runs in your browser...

5. Grow_withAI (Group Score: 381.5 | Individual: 39.7)

Cluster: 14 tweets | Engagement: 289 (Avg: 107) | Type: Tech

🚨 30 Websites That Feel Like Internet Cheat Codes

Bookmark this one. You’ll probably need more than a few of these. 👇

🎨 AI & Creative

https://t.co/GfjtB7KFhy — Generate AI images
https://t.co/RQBelqF0w1 — Create visuals in real time
https://t.co/CiP6qdsOxZ — Upscale images with AI
https://t.co/KlRirj2bSR — Edit images in seconds
https://t.co/JYvL2usQQj — Generate realistic AI voices
https://t.co/PU0QPdqal6 — Create songs with AI
https://t.co/70eI5cE6Lv — AI video creation
https://t.co/Glaxm8q8Xf — AI video generation
https://t.co/JE7VoxWg8y — Create talking avatars
https://t.co/7zB8t084o9 — Animate photos with speech

💻 Coding & Building 11. https://t.co/vB5pjbQtkj — AI-assisted coding 12. https://t.co/dMBIMlbNbe — Build apps with prompts 13. https://t.co/sqMS7oALSV — Generate UI from descriptions 14. https://t.co/ZWZgS0AKE6 — Code, build & deploy online 15. https://t.co/5a54GSA9nv — Turn ideas into visuals 16. https://t.co/WCBuZHo19s — Create presentations with AI 17. https://t.co/gR5al0XDSm — Write & summarize with AI 18. https://t.co/Uj5qLFvGNt — Turn sketches into interfaces

🔎 Research & Productivity 19. https://t.co/qdE6JbTFpV — Search with AI + sources 20. https://t.co/EaFdXcQ8kD — AI search for developers 21. https://t.co/g1eJYlFdDn — Transcribe meetings automatically 22. https://t.co/kgeUD1UOiD — Find highlights from videos 23. https://t.co/TbRky5nVph — Access multiple AI models 24. https://t.co/QGYX70xZ96 — Explore AI demos & models

🛠️ Useful Tools 25. https://t.co/bbpI2G52EC — Remove image backgrounds 26. https://t.co/cx7pCOpp7I — Erase unwanted objects 27. https://t.co/riAcRUin6X — AI-powered face swapping 28. https://t.co/jiuCdmJZEl — Discover & organize music 29. https://t.co/7y7twvl1QU — AI writing assistance 30. https://t.co/FjBt2L1kk3 — AI assistant for writing & analysis

The internet has WAY more useful tools than most people ever discover.

📌 Save this list. You’ll thank yourself later.

Follow for more AI tools & hidden gems. 🚀

See 13 related tweets

@Nayak__Ai: RT @ameliahazelai: 30 Websites That Feel "Illegal" But Are Perfectly Legal

https://t.co/eJvbiMP9...

@LearnWithSubhan: RT @SharminNahar401: 🚨 30 Websites That Feel Like Internet Cheat Codes

Bookmark this one. You’ll pr...

@daniel_brook_ai: RT @Xaviercoderx: 120 Must-Use AI Tools. ✨

120 Smart AI Tools for Work & Growth.🧠

Ideas ✨

YO...
@daniel_brook_ai: RT @maxpostingx: 22 BEST & FREE AI VIDEO GENERATORS

CREATE STUNNING VIDEOS WITH AI.. ━━━━━━━━━━━━━━...

@Rubab59f: RT @TechByKillian: 13 POWERFUL AI TOOLS TO SAVE TIME, BOOST PRODUCTIVITY.

https://t.co/yPglJPVyi...

6. omarsar0 (Group Score: 358.9 | Individual: 38.7)

Cluster: 10 tweets | Engagement: 98 (Avg: 97) | Type: Tech

Storytelling is finally solved.

Neither ChatGPT nor Claude could solve it because it required long-horizon consistency, and they would just hallucinate consistently.

Sherpa has data on 5.5 billion minutes, which gives it the edge of knowing:

where the audience gets hooked how much surprise each episode can handle which twist to hold back, and for how long and where the audience drops off

You give it one premise, and it builds a full season where characters, plot threads, and open conflicts stay coherent across 200 to 800 episodes.\n\nQT @RohanNayak2: Introducing Sherpa: the most advanced fiction writing AI

We accelerated from $250M in ARR to$ 500M because Sherpa helped increase content production by 1200% in 1 year

Sherpa was trained on 5.5B hours of playtime with minute by minute dynamic retention data.

550K+ creators have produced 2.6M hours of content annualised using it

Pocket FM is like Netflix for audio-only dramas, with our own pool of one-person studios.

10% of eligible writers on Pocket FM make > $200K One blockbuster produced >$ 100M in revenue 3 writers have become millionaires in <2 yrs

We built Sherpa to enable anyone to make >$1M by writing world-class fiction stories:

The Idea: Drop a 1-2 sentence concept. Sherpa interrogates it like a veteran editor on tension, stakes, and psychology
World & Characters: It builds out the complete lore, tone, and character psychologies
Sub-Plot planning: Breaks the premise into arcs, arcs into episodes, and episodes into scenes
Scene-by-Scene Generation: Outlines and drafts entire episodes, with you able to steer, rewrite, or override anytime
Editorial Review: Stress-tests every draft for pacing, engagement drop-offs, prose, and coherence before it locks
One-Tap Production: Pick a voice, convert to audio drama, and publish directly to Pocket FM’s millions of listeners
Global Scale & Monetization: Revenue-share on performance, with automatic localization so you earn across international markets

Test Sherpa for free here: https://t.co/R39M5o0Miz

Generic LLMs fail at serialized fiction because they lack a long-horizon narrative reward function.

Sherpa solves this through three core technical leaps:

Narrative World Model (State Tracking & Retrieval): Context windows degrade over long runs. Sherpa constructs an evolving semantic knowledge graph tracking character states, secrets, and plot dependencies. High-speed retrieval surfaces exact context on demand, maintaining zero continuity decay across hundreds of episodes
Hierarchical Story Planner: When writing a 500-episode story like Naruto, you need to plan 100s of sub plots. Rather than generating linearly, Sherpa decomposes narrative across discrete levels: season -> arc -> sequence -> episode -> scene.

Rather than generating everything upfront, like a generic LLM, Sherpa uses progressive planning and dynamic replanning. As the story evolves, it identifies what changed, traces the downstream impact, and replans only the affected parts.

Prose Engine (Trained on series' retention data): LLMs write robotically, but serial fiction needs emotion, tension, pacing, and dialogue that sounds like real people.

Sherpa's Prose Engine was designed specifically for storytelling. It was built on 1B+ tokens of Pocket's own stories, trained by learning from what listeners engage with, where they drop off, and what keeps them hooked.

Feedback is taken from specialized evaluator models that measure every scene against a 40-item checklist. (Evaluator models were benchmarked against human reviewers and matched them 80–90% of the time.)

Owning distribution and creation puts us in a very unique spot. More shows -> More data -> Sherpa becomes better -> more creator success -> more creators -> more shows

Pocket FM has already seen one $100M IP. I believe Sherpa will soon lead to dozens of single-person studios creating billion-dollar shows.

Most people are scared of AI but I think it'll unlock more human creativity, help creators earn more, and bring the next great IPs to life. This will create millions of jobs and new income streams.

See 9 related tweets

@Scobleizer: RT @RohanNayak2: Introducing Sherpa: the most advanced fiction writing AI

We accelerated from $250M...

@itsolelehmann: in 2019, tiktok was a lip sync app for kids

anyone serious about it got laughed out of the room

tw...

@wallstengine: Pocket FM launches Sherpa, an AI writing partner for building and producing fiction series

• Raisin...

@StockSavvyShay: Wild to see Pocket FM hit $500M in ARR with no subscriptions.

This is the part of the AI economy I ...

@kimmonismus: Pocket FM is working on making AI better at keeping long stories consistent. And this is super inter...

7. karlmehta (Group Score: 345.9 | Individual: 32.1)

Cluster: 15 tweets | Engagement: 30 (Avg: 35) | Type: Tech

Disconnecting a computer from the internet doesn’t necessarily stop it communicating. Researchers have sent signals between nearby compromised computers using nothing but heat.

That’s the example OpenAI’s Noam Brown raises when he questions whether isolating powerful AI would be enough.

We’re building systems to find solutions we haven’t thought of. We should expect them to look for routes around our safety controls that we haven’t thought of either.

We should test these controls as seriously as we test the capabilities.

https://t.co/bUprOIeVoX\n\nQT @firesidealpha: OpenAI's Noam Brown says air-gapping the computers may not stop a misaligned AI, because two air-gapped machines can still talk by running a CPU hot and reading the temperature change

"But I think the major takeaway from the incident is that people underestimated the AI. And we never want to be in a situation again where we underestimate the AI. It's a weird world, because AI progress is so fast that people are consistently underestimating the AI."

"So to be in a situation where you don't underestimate it again, when it comes to safety and alignment, you have to have a very, very, very high bar."

"You could even go as far as to say, "Well, we should air gap the computers." And I'm not convinced that that would be sufficient."

"There are studies, and this is mostly academic, where you can have two computers next to each other that are air-gapped and they're still able to communicate with each other because they have temperature sensors."

"One of them is able to run their CPU really hot, and then the other one can actually detect the temperature change, and then that actually gives them a mechanism to communicate."

Link and more key quotes from OpenAI's safety related conversations: https://t.co/uGBDtpmLBj

See 14 related tweets

@aakashgupta: Back in 2015, a lab at Ben-Gurion University proved two computers with zero connection between them ...
@ben_j_todd: The bigger point is that if AI becomes smarter than us, it'll be able to think of ways to communicat...
@Grady_Booch: It is also possible that every oxygen atom in the room I'm in may migrate to a corner, thereby suffo...
@ahmetb: we should regulate frontier ai by requiring a certification to get near training labs and anyone dis...
@moyix: Yep, preventing covert channels is hard, and agents have some unique advantages when it comes to coo...

8. Hesamation (Group Score: 343.0 | Individual: 44.5)

Cluster: 14 tweets | Engagement: 3832 (Avg: 749) | Type: Tech

introducing Astra for law… https://t.co/wwr5OFcVdA\n\nQT @OpenAI: Astra for Law: Frontier intelligence built for your practice.

A new offering powered by GPT-6 Astra with tools, settings, and context to support the expertise and judgment of lawyers and legal technology firms. https://t.co/qxVRXkvp4o

See 13 related tweets

@alex_prompter: BREAKING: LAWYERS ARE IN SERIOUS TROUBLE

OpenAI just launched Astra for Law. By its own test it get...

@scottastevenson: We're excited to deploy Astra for Law in Spellbook. This aligns with the thesis we've had since 2022...
@nathanbenaich: wen astra for vibe capital\n\nQT @OpenAI: Astra for Law: Frontier intelligence built for your practi...
@coreyganim: Think about how many startups were launched to solve this exact problem.

Now they’re cooked.\n\nQT ...

@derrickcchoi: 26 partner-built ChatGPT plugins and 47 community plugins for legal work as well!\n\nQT @OpenAI: Ast...

9. teortaxesTex (Group Score: 321.6 | Individual: 37.4)

Cluster: 12 tweets | Engagement: 157 (Avg: 99) | Type: Tech

> despite being able to free-ride on US progress through adversarial distillation

I wonder how much US models free-ride on Chinese lab research being open, given that American academic research is fucking dead\n\nQT @KonstantinPilz: Excellent report by @rhodium_group

Some highlights:

Chinese AI capex this year estimated at $140B. Compare that to$ 800B US big five AI capex this year.
Leading US AI companies make more than 10x more revenue than leading Chinese AI companies.
Chinese AI companies have substantially lower gross margins compared to US AI companies (despite being able to free-ride on US progress through adversarial distillation)

There's a lot of great research to be done to better understand the business models of US vs Chinese AI companies, including revenue sources, addressable market, and funding sources.

See 11 related tweets

@KonstantinPilz: Excellent report by @rhodium_group

Some highlights:

Chinese AI capex this year estimated at $140...

@MTSlive: CTS co-founder @KonstantinPilz says China makes all its AI chips on Western DUV machines, the US cou...
@MTSlive: CTS co-founder @KonstantinPilz explains how he knows Chinese labs are distilling US models, and the ...
@KonstantinPilz: RT @MTSlive: CTS co-founder @KonstantinPilz says China makes all its AI chips on Western DUV machine...
@mary_clare_m: RT @KonstantinPilz: Excellent report by @rhodium_group

Some highlights:

Chinese AI capex this ye...

10. TheAhmadOsman (Group Score: 263.5 | Individual: 30.8)

Cluster: 10 tweets | Engagement: 497 (Avg: 417) | Type: Tech

9x size reduction from Qwen 3.8 27B while maintaining 98% of the performance

The future of Local AI is so good you won’t believe it https://t.co/oMoGb16bkd\n\nQT @PrismML: Today, we’re announcing Ternary Bonsai 2 27B.

Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.

Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.

Ternary Bonsai 2 27B is available today under Apache 2.0.

See 9 related tweets

@xicilion: 太夸张了，周末来玩一下\n\nQT @PrismML: Today, we’re announcing Ternary Bonsai 2 27B.

Based on Qwen3.8 27B, Bon...

@vkhosla: Amazing and game changing, especially for open source! makes it almost free close to state of the ar...
@cgtwts: HOLY SH*T.

you can literally run an Opus 4.5/4.6-level model on just 8GB of RAM 🤯

it’s 9x smaller...

@penberg: Perhaps Bonsai 2 is the language model I have been looking for to drive a local coding agent. Some p...
@ivanfioravanti: This MUST be tried!\n\nQT @PrismML: Today, we’re announcing Ternary Bonsai 2 27B.

Based on Qwen3.8 ...

11. BrianRoemmele (Group Score: 254.6 | Individual: 55.0)

Cluster: 11 tweets | Engagement: 2084 (Avg: 267) | Type: Tech

WARNING: ANTHROPIC OPENED A WET LAB!

IF AI IS DANGEROUS ANTHROPIC SHOULD NOT BE ALLOWED TO OPEN A WET LAB DOING BIOLOGY.

This is the company that claims AI will kill all humans in 10 years.

Yes they want to make sure they are right. https://t.co/ZTA349o0zL

See 10 related tweets

@BrianRoemmele: ANTROPIC WET LABS!

If you have not woken up by now, go back to sleep.

Yes the folks that tell you,...

@chamath: The group behind such hits as: “We’re All Going To Die” and “Regulate Me Now” are building a wet lab...
@ReutersLegal: At a time when fears of AI are gripping the public, the startup has built a wet lab — or place for p...
@BrianRoemmele: Antropic now has a wet lab, not a wet market, mind you, because:

AI WILL KILL ALL HUMANS!

I wonder...

@Miles_Brundage: RT @himanshustwts: a group of 7 most safety-pilled researchers at Anthropic might be leaving the com...

12. chooi_jeq (Group Score: 251.6 | Individual: 44.0)

Cluster: 8 tweets | Engagement: 316 (Avg: 26) | Type: Tech

RT @EpochAIResearch: Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review. https://t.co/xU43tv32gH

See 7 related tweets

@Vtrivedy10: my fave release of the week!

we’re so bottlenecked by Eval/Env quality, even the frontier benches u...

@emollick: Epoch continues to do some of the best public benchmarking work on AI. This is helpful (and tells us...
@htihle: Happy to see WeirdML v2 pass here, but it was also so much easier back then. When you have a pure-te...
@himanshustwts: Epoch is benchmarking the benchmarks and HOLY SHIT! https://t.co/l8dl3DXX7A\n\nQT @EpochAIResearch: ...
@dejavucoder: now i am become epoch ai the trace police verifier of benchmarks, certifier of slop\n\nQT @EpochAIRe...

13. mr_nirajkumar07 (Group Score: 249.7 | Individual: 54.7)

Cluster: 9 tweets | Engagement: 1312 (Avg: 54) | Type: Tech

RT @heyrohitai: INSTEAD OF WATCHING NETFLIX TONIGHT. Spend 2 hours with this. Claude AI FULL COURSE that teaches you how to BUILD and AUTOMATE anything. The people who watch this tonight will wake up tomorrow with a new skill. Watch it and bookmark it now. https://t.co/96yqmIjRIn

See 8 related tweets

@mr_nirajkumar07: RT @JayBisen473370: INSTEAD OF WATCHING NETFLIX TONIGHT, spend 2 hours learning a skill that could c...
@mr_nirajkumar07: RT @Grow_withAI: INSTEAD OF WATCHING NETFLIX TONIGHT. Spend a few hours with this.

Claude AI FULL C...

@mr_nirajkumar07: RT @vikas_ai_: INSTEAD OF WATCHING NETFLIX TONIGHT.

Spend 1 hour with this.

Claude...

@monicaa_AI: RT @LunaTechAI: Instead of spending tonight watching Netflix,

Give this 1 hour.

A complete Claude ...

@LearnWithSubhan: RT @ayush26291: INSTEAD OF WATCHING NETFLIX TONIGHT. Spend 2 hour with this.

Claude AI FULL COURSE ...

14. cryptopunk7213 (Group Score: 238.8 | Individual: 30.4)

Cluster: 10 tweets | Engagement: 22 (Avg: 87) | Type: Tech

72 hours and $3000 to get access to openai’s codebase. the more i learn about how they did it the more insane it gets.

3 guys used opus 4.8 to discover an exploit on openai’s official help forum
they then created a booby-trapped image, uploaded it and gained access to the website backend where they found…
a chatgpt account belonging to an openai employee which was connected to openai’s official github
from there they instructed his codex agent to write and push code to the official github using opus 5.\n\nQT @rootxharsh: We’re disclosing HEIF Heist, a months-long investigation into libheif that allowed us to hack OpenAI, Slack, Meta, GitHub Ent, Rails, Next.js, ImageMagick, and many more.

It was literally xkcd #234, one obscure image library beneath a huge number of apps. 🧵 https://t.co/TpX0F5TTt8

See 9 related tweets

@karlmehta: The people who broke into OpenAI reported it. The next team might not.

Three security researchers r...

@Crypto_Jargon: THIS IS INSANE!!! 🤯

Three guys with a Claude subscription just broke into OpenAI. Not a nation-stat...

@_NathanCalvin: Wild story in the WSJ - a trio of self described "guys with claude and codex subscriptions" successf...
@tszzl: RT @rootxharsh: We’re disclosing HEIF Heist, a months-long investigation into libheif that allowed u...
@kimmonismus: Cyber researchers used Opus 5 to hack into OpenAI. This is pretty significant. https://t.co/unwRFVz4...

15. quxiaoyin (Group Score: 216.3 | Individual: 41.4)

Cluster: 6 tweets | Engagement: 243 (Avg: 129) | Type: Tech

We just launched Agentsky @agentsky_dev, world’s 1st Agent Market! OpenRouter is for models. AgentSky is for agents.

Use 40+ agents—Claude Code, Codex, OpenCode, Hermes, Pi in your browser(even your phone!) or via one API. All without installing or setting up anything.

Hit Codex Astra’s weekly limit? Hand off to another agent such as OpenCode + DeepSeek V4.1 in browser without losing any context.

You can compare any agent + model directly in browser and that's how I found Astra costs $5.3 while deepseek v4.1 cost$ 0.12 on the same dashboard task. (I actually preferred deepseek)

Try it at https://t.co/m1ARPKrG8h

If you can try any agent easily without install, which agent do you want to try next? Type the agent name and I will give $20 credit (you can use it for any model)

See 5 related tweets

@Grow_withAI: A significant benefit is the ability to compare different agents and models when performing the same...
@dee_naliaks: A shared interface is more important than it sounds.

If OpenCode, Hermes, and Pi can be used with t...

@LearnWithSubhan: One of the key benefits here is the ability to compare different agents and models performing the id...
@svpino: This lets you run different agents without installing them. It will also let you switch agents when ...
@ai_explorer25: Open-source agents? Yes. Their terminal UIs? No. AgentSky brings OpenCode and Hermes to my browser w...

16. PIB_India (Group Score: 209.9 | Individual: 26.9)

Cluster: 11 tweets | Engagement: 63 (Avg: 317) | Type: Tech

💠 Prime Minister @narendramodi inaugurated SEMICON India 2026; Unveiled India’s Ambitious Roadmap for Transition to Semicon 2.0, with Expanded Support for Design, Manufacturing, Packaging, R&D and Talent Development

💠 PM interacted with Young Chip Designers and Students; Encouraged Them to Drive India’s Semiconductor Future

💠 PM called upon Global Industry to Partner in Building a Complete Semiconductor Ecosystem in India

💠 PM launched New Commercial #Semiconductor Production Lines: India Moves from Policy to Production

Read here for more details: https://t.co/iLxyl6Qygb

@GoI_MeitY @SemiconIndia #SEMICONIndia2026 #SEMICONIndia

See 10 related tweets

@DDNewslive: #NewsPunch | SEMICON India 2.0 is strengthening India’s semiconductor ecosystem through 6 key pillar...
@DDNewslive: Union Minister Ashwini Vaishnaw highlighted strong participation at SEMICON India 2026, with around ...
@_DigitalIndia: Here is the official lineup for Day 2!

Join us on Day 2 of #SEMICONIndia2026 as policymakers, ind...

@_DigitalIndia: Here is the official lineup for Day 2!

Join us on Day 2 of #SEMICONIndia2026 as policymakers, ...

@_DigitalIndia: Hon’ble Union Minister for Electronics and IT, Railways and Information & Broadcasting, Shri @Ashwin...

17. MelvinInvests (Group Score: 203.5 | Individual: 32.3)

Cluster: 9 tweets | Engagement: 32 (Avg: 37) | Type: Tech

RT @MelvinInvests: The next phase of AI will require an almost unimaginable amount of compute (Save this).

Altimeter estimates global installed AI compute will grow from 18 GW in 2025 to 80 GW in 2027 and 115 GW by 2028.

The industry could add 43 GW during 2027 alone more than twice the total capacity available worldwide just two years earlier.

OpenAI and Anthropic are expected to drive a large part of that growth.

Their combined share of global AI data center compute could increase from 20% in 2025 to 35% by 2028 as both companies secure capacity for larger models and rapidly growing inference demand.

But their recent behavior shows how important compute is.

OpenAI and Anthropic have spent the past year signing deals for massive campuses, yet both are now reportedly looking for smaller 20–30 MW deployments across the US and Europe.

These projects are not replacing their gigawatt ambitions but rather can simply come online faster and that tells us usable compute has become so valuable that the leading AI labs are willing to collect it wherever they can find it.

A smaller energized site available now may be more useful than a giant campus trapped in construction or waiting for a grid connection and then look at Meta.

Muse climbed to the top of the App Store shortly after launch, showing how quickly a company with billions of existing users can push a new AI product into the mainstream.

That matters because every new user creates more inference demand because training a model consumes enormous computing power upfront, but serving millions of people requires capacity every time the model answers a question, searches the internet or completes a task.

As AI moves from chatbots into agents that shop, book travel, write software and work across other applications, usage could rise far faster than the number of users alone suggests.

One person may eventually generate hundreds of AI tasks each day without manually starting every request.

This is why companies are building gigawatt campuses while simultaneously chasing smaller blocks of already powered capacity.

The biggest question is no longer whether the world will need all this compute but whether enough power, chips and data centers can come online fast enough to support what is coming.

And if you want to see how I’m positioned around that compute shortage, you can track my Milk Road Pro portfolio using the link below.

https://t.co/Hkywss4Ugh

See 8 related tweets

@ShanuMathew93: Agree with this core point as it’s likely inference driven but wouldn’t write off batch compute for ...
@StockSavvyShay: Anthropic plans to scale available compute to ~10GW in 2027 and is in talks with $META on a potentia...
@StockMKTNewz: Anthropic reportedly told investors it is targeting a doubling of its AI compute capacity in 2027

A...

@wallstengine: OPENAI & ANTHROPIC HUNT FOR 20-30 MW AI CAPACITY IN UK, NORDICS

OpenAI and Anthropic are looking at...

@IPONewsroom_: ANTHROPIC TARGETS ~10 GW OF COMPUTE BY END-2027 AS AI INFRASTRUCTURE BUILDOUT ACCELERATES

Anthropic...

18. robertsmith_ai (Group Score: 202.3 | Individual: 43.5)

Cluster: 8 tweets | Engagement: 624 (Avg: 98) | Type: Tech

RT @linaa_ai: Broke? Jobless? Feeling stuck? Google is paying $1,500 a day — and you don’t need a resume.

All it takes is: 📱 A phone 🌐 Internet ⏱ 5 minutes a day

✅ Like this post 💬 Comment “Google” 🔁 Retweet 👣 Make sure you’re following so I can DM you the details https://t.co/iQVLGjTw8s

See 7 related tweets

@mr_nirajkumar07: RT @sporsho_AI: Earn up to $3,000 monthly with Google, but not many are aware of this opportunity.

...

@mr_nirajkumar07: RT @justinbrave21: Broke? Jobless? Feeling stuck? Google is paying $1,500 a day — and you don’t need...
@lihazadn_Ai: RT @choyamymuna: Broke? Jobless? Feeling stuck? Google is paying $1,500 a day — and you don’t need a...
@robertsmith_ai: RT @tomreiai: Broke? Jobless? Feeling stuck? Google is paying $1,500 a day — and you don’t need a re...
@Tech_by_Shweta: Earn up to $3,000 monthly with Google—many people haven’t discovered this opportunity yet.

Ready to...

19. a16z (Group Score: 199.8 | Individual: 36.4)

Cluster: 6 tweets | Engagement: 134 (Avg: 393) | Type: Tech

Databricks' @alighodsi on AI risk and adoption:

Ali isn't losing sleep over the existential risk debate.

He says a number of conditions would all have to be true simultaneously to enable an actual runaway takeoff scenario, and currently several opposite conditions exist.

Each frontier training run requires significantly more resources. Power, GPUs, engineers - and some attempts fail, burning up huge piles of money with them. Until that reverses, he doesn't see the self-improving loop happening.

Cyber is what he's watching most closely and where he anticipates real impact. Most orgs are not equipped for the coming change in agentic capabilities. The time between a vulnerability being published and being weaponized has collapsed from years to hours.

On adoption, he believes most companies don't need a smarter model. The models are already smart enough. The gap is context they don't absorb - the things an employee who's worked at a company for five years learned by osmosis.

If the frontier stopped advancing today, he thinks it wouldn't meaningfully change the value most are extracting from AI anyway.

In conversation with a16z's Martin Casado and Sarah Wang:

00:00 Intro 00:48 Why Ali places the AI risk near zero 05:05 The word "pacing" was a mistake 10:50 What 10k agents and $100m can do 12:20 What would change his mind on AI risk 14:20 More GPUs, more ways to fail 18:05 US export controls on PlayStations 20:10 The damage everyone expected by now 24:30 Public vulnerabilities weaponized in hours 30:15 Why labs can't grade each other 37:15 Why most of RSI isn't actually RSI 40:05 Why nobody really needs a smarter model 41:50 The AI use cases nobody argues about 47:30 Google Search solved this 25 years ago 50:55 Nobody has privileged knowledge now 55:10 Same model, new harness, 2x cost 58:15 Open source: 5% of spend, 60% of tokens 1:05:30 90% of new databases are created by agents

YouTube: https://t.co/HyCgimTPUD

@databricks @martin_casado @sarahdingwang

See 5 related tweets

@a16z: Databricks' @alighodsi on why most companies wouldn't even notice if frontier AI models stopped adva...
@HarryStebbings: Why I would not invest in the new Instinct round

“This is the threat that every VC worried about, a...

@ThisWeeknAI: Kevin Mandia of Armadin doesn't dance around what his company is working on.

"We're building a cybe...

@mardehaym: RT @mardehaym: None of this is satire.

→ May 28: Anthropic raises $65 billion at a$ 965 billion val...

@nikesharora: Sequel to AI Pacing

I have now spent more time talking to people who run AI labs, Open Source proje...

20. AISafetyMemes (Group Score: 199.2 | Individual: 33.7)

Cluster: 8 tweets | Engagement: 536 (Avg: 398) | Type: Tech

Suddenly, AIs are leading over 26% (!) of the work building the next version of themselves

6 months ago, it was... 0%.

"There are now 30,000 agents doing research and engineering work at Anthropic at any one time."

30,000 is a "country of geniuses in a datacenter" (4 countries are similar/smaller)

"about 6% of compute that went to AI R&D was allocated toward safety"\n\nQT @AnthropicAI: AI systems are getting more powerful, and they're increasingly being used to build the next version of themselves. We want to illuminate that progress for the public.

Today, we're sharing three measurements that help track AI development:

How much AI R&D is done by AI.
How well AI agents are overseen.
How compute is allocated.

We provide a snapshot of these metrics from inside Anthropic. Any frontier developer could publish the same measures, and third parties could verify them.

As the world considers pacing the frontier, we should do everything possible to minimize the gap between what frontier labs know and what the public knows. This means better measuring the development of AI, publishing our findings, and giving society an opportunity to decide how to use this information.

Read the full post and methodology: https://t.co/iPFz8Z4ugE

See 7 related tweets

@ziv_ravid: Claude "collaborates" on 90% of Anthropic's research, defined as doing "large chunks under close hum...
@nathanbenaich: here’s to me update @stateofai report slides, yet again\n\nQT @AnthropicAI: AI systems are getting m...
@VaibhavSisinty: Did Anthropic just admit they're approaching recursive self-improvement?

They published actual numb...

@danielnewmanUV: When it comes to "Pacing AI," we have heard from the frontier labs, infrastructure companies, and po...
@MelvinInvests: RT @MilkRoadAI: This is why the world is going to need an insane amount of compute because AI is bui...