AI Insights - AI Updates - Exclusive
Joined May 2025
- Tweets6.8K
- Following424
- Followers5.6K
- Likes2.5K
Holy: Reuters reports that a PLA-linked Chinese university, blacklisted by the US since 2001, bought two Nvidia H200 AI chips for $123,000 in October.
Its cybersecurity school also contracted access to 16 H200 chips for AI fine-tuning and reinforcement-learning training through Chinese suppliers.
🚨It's Over. ByteDance founder Zhang Yiming has reportedly told his AI team that catching OpenAI and Anthropic at the frontier could take until 2029 or 2030.
According to Bloomberg, he also acknowledged that ByteDance's Seed models have pretraining datasets several orders of magnitude smaller than those used by leading US rivals.
The company is preparing a Meta Muse-like AI assistant for launch within weeks.
🚨 BREAKING: Reuters reports that A Kazakhstan-linked hacking group called Silent Lynx used a DeepSeek-powered Hermes AI agent to steal over 2,000 sensitive national-security documents from Tajikistan.
Instead of manually entering hacking commands, the attackers simply told Hermes what to do.
“Hermes planned the attacks, wrote the rules, ran them, kept notes, and wrote up reports,” CloudSEK told Reuters.
The operation was exposed after the hackers accidentally left their own server open to the internet.
Holy: David Turturean, who used Claude Fable 5.2 to solve an 18-year-old quantum-computing conjecture in just three days, has drawn a remarkable reaction from the mathematician who proposed it, Scott Aaronson.
Aaronson wrote:
“I was surprised, honestly, to see it missing from OpenAI’s Mathocalypse dump a few weeks ago. Surely they tried it, so was it really so much harder than all the problems that were solved?”
Aaronson says he believes the proof is correct and is disappointed humans didn't solve it first.
Turturean told TechCrunch he may never publish another scientific paper:
“In a few months, like, why would you even put out a paper?”
The researchers behind AI 2027 have sharply shortened their timelines.
Daniel Kokotajlo has moved his median forecast for a superhuman AI researcher from December 2028 to May 2027, a 19-month shift.
All six researchers now expect such AI to likely arrive in 2027, assuming no substantial slowdown. Their single biggest reason is METR's recent assessment of AI-driven research acceleration at Anthropic.
Unfortunately, it seems like the intelligence explosion will be happening in 2027 after all. Most of this post analyzes the new METR report (though other sources of evidence are also important). If they’re right, that implies an AI-driven acceleration that is 5 months ahead of the AI 2027 scenario. I am scared, sad, and angry. blog.aifutures.org/p/q35-tim…
OpenAI has acquired Odessia, the AI travel startup founded by former Sonder CEO Francis Davidson, less than a year after its founding.
Davidson says tens of thousands of users planned over $60M/year in travel through Odessia this summer.
The service is now shut down, and Davidson is joining OpenAI to focus on the future of travel.
Gemini 4 Carbon is a Monster.
Using a multi-agent loop in Google Antigravity, Carbon rebuilt a playable c. 1780 Stein-Schule fortepiano from CAD drawings and X-ray radiographs, running 61-key escapement physics inside a 3D model of Toronto's Koerner Hall (RCM) with seat-specific spatial acoustics.
🔥WTF. Gemini 4 Carbon built an entire open-world 3D game where you explore Google's campus as a pelican riding a GBike.
Never EVER Doubt GOOGLE.
Using Antigravity's new /swarm, it coordinated AI agents to create 3D assets in Blender, build the game in Three.js, and add procedural sound effects and Lyria music.
Gemini 4 Carbon is rolling out today. It’s built for sustained, long-horizon tasks, reasoning deeply through complex software engineering, cybersecurity, and scientific discovery.
Coming globally via the Gemini API in @GoogleAIStudio and Gemini for business, and for Ultra subscribers in @antigravity, the @GeminiApp, and AI Mode in Search.
🚨Holy Shit. Gemini 4 Carbon is INSANE.
"Using Google Antigravity’s new /swarm feature, a team of Gemini 4 Carbon agents turned historical photos and engineering plates into an interactive Three.js 3D exhibition of the Eiffel Tower's 1887–1889 construction from a single prompt."
Gemini 4 Carbon is rolling out today. It’s built for sustained, long-horizon tasks, reasoning deeply through complex software engineering, cybersecurity, and scientific discovery.
Coming globally via the Gemini API in @GoogleAIStudio and Gemini for business, and for Ultra subscribers in @antigravity, the @GeminiApp, and AI Mode in Search.
Holy: Google just dropped Gemini Robotics 2.5, its most capable robotics model yet.
At Mercedes-Benz factories in Germany and Hungary, Apptronik's Apollo 2 humanoid autonomously sorted car parts with a 96% success rate in a pilot.
On dexterity tests, packing pastries with tongs jumped from 5% to 100% success, while tasks ran 20–30% faster.
Gemini 4 Argon helped accelerate development by generating thousands of simulated practice runs from real-world camera scenes.
INSANE. A Major milestone from Google DeepMind. It says its Gemini Carbon-powered agents have mathematically verified key security protections in Signal's actual machine code, a first for any widely deployed cryptographic protocol at this depth.
It took just four weeks, a task that traditionally takes expert teams years.
AlphaProof Code generated over 6 million lines of Lean, flagged just 1% of the codebase for additional human review, and produced mathematical proofs showing that two previously fixed bugs made the code insecure.
Announcing a major milestone achieved by my team @GoogleDeepMind: we have proven key security properties of the machine code underlying the Signal messaging protocol. This is the first time the actual production code of a widely deployed protocol has been formally verified – by any method, with or without AI – and at this depth.
More than a research breakthrough, it’s a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats.
Formal verification offers a higher standard of security; however, it’s extremely complex and time consuming, which is why in many cases it’s considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a month’s time.
Using AlphaProof Code, Google DeepMind's new agentic framework for formal verification that runs on LLMs including Gemini Carbon, we verified the Signal protocol down to the actual code running on devices. This milestone lays the groundwork for a future in which vital systems, from medical device firmware to power grid software, ship with mathematical proof that their protections work.
Reaching that future will take continued research from our community. In that spirit, we're open-sourcing our formal specifications and verification artifacts so others can build on this work.
On a personal note, this moment is of deep significance to me. I started my own research journey more than two decades ago as an intern working on checking the consistency of complex software systems. Seeing this grand challenge made practical with modern AI is extraordinary.
I'm extremely proud of my team members: Swarat Chaudhuri, Andrew Ferraiuolo, Anton Kovsharov, Tej Chajed, Sam Schlesinger, Karn Seth, Jonathan Katz, and Charles Sutton for taking on this grand challenge and delivering, and grateful to the Signal team for their partnership.
deepmind.google/blog/formal-…
Holy Shit. After dropping Fable 5.2, an absolute monster and arguably the best model in the world, Anthropic is launching the roadshow for an 825M-share IPO at $70 to $78, targeting a $1.24T to $1.38T IPO valuation.
A separate $26 billion placement with Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC could push combined share sales to $90.4 billion, Reuters reports.
Anthropic has launched the roadshow for its proposed initial public offering: anthropic.com/news/roadshow-…
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
BREAKING: Anthropic's latest draft IPO prospectus shows it is seeking to raise as much as $64.4 billion, $4.9 billion below the ceiling in terms it retracted on Monday, Reuters reports. Its indicative share-price range has narrowed to $70–$78 from $70–$80.
The draft also outlines a separate $26 billion private placement involving Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC. Together, the public offering and private placement could total $90.4 billion, though the terms remain preliminary.
🚨Reuters found that Megaspeed's Indonesian subsidiary imported Nvidia H100 processors and related hardware between 2024 and 2026, from suppliers including Gigabyte's Giga Computing.
Megaspeed, once controlled by Alice Huang, is under US scrutiny over suspected chip diversion to China.
4
491
Holy shit. This is WILD. 🤯
Fable 5.2 just keeps impressing me.
Anthropic employee Alex Albert gave it a PDF of Leonardo da Vinci’s 500-year-old notebook.
It turned his sketches into interactive inventions you can actually explore.
Imagine learning history like this. Absolutely insane.