Must attend if possible

I'm interested in more detailed work on how much profit AI companies will capture and whether there will be a bust in the next 2y. Appreciated the emphasis that "market crash" != "low-impact technology"

Another beautiful paper. 👏 Shows how AI can automatically discover and test better systems algorithms. Across tasks, some evolved algorithms run 5x faster or cut cost by 26%. The method, called AI Driven Research for Systems, ADRS, generates code, runs it, scores it, then

BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2025 #NobelPrize in Chemistry to Susumu Kitagawa, Richard Robson and Omar M. Yaghi “for the development of metal–organic frameworks.”

Excited to share early results about CodeMender, our new AI agent that automatically fixes critical software vulnerabilities. AI could be a huge boost for developer productivity and security. Amazing work from the team - congrats!

Really excited to finally announce CodeMender! As part of this we've already submitted and upstreamed several patches to OSS projects via OSS-Fuzz. Check out our post at: https://t.co/qgnroQyIzN There will be more technical details and exciting announcements to come!

John Clarke, UC Berkeley emeritus professor, awarded the 2025 #NobelPrize in #Physics. The Nobel Prize committee honored Clarke "for the discovery of macroscopic quantum mechanical tunneling and energy quantization in an electric circuit."

The term 'AGI' is a problematic and shifting goalpost, and focusing on a single 'AGI event' is less productive than focusing on the continuous development and measurement of an AI's generality across an expanding set of domains.

Confused on multiple dimensions: 1. I am not at all anti-AGI. have said repeatedly that I see considerable potential value for AGI, especially for scientific discovery. I just don’t think AGI is close, and I don’t think we should rush to build it until we have some clue how to

Introducing SkyRL tx 🧸, an open-source project to implement the Tinker API. The SkyRL team is excited about the Tinker API and the opportunities of using a single canonical interface that unifies training and inference. SkyRL tx lets you run a Tinker-like service locally today,

Training our advisors was too hard, so we tried to train black-box models like GPT-5 instead. Check out our work: Advisor Models, a training framework that adapts frontier models behind an API to your specific environment, users, or tasks using a smaller, advisor model (1/n)!

Proud to be part of the team!

Following up on the AlphaEvolve code opt. agent, I am happy to share how our team at @GoogleDeepMind has developed the CodeMender agent to design/apply patches to fix security vulnerabilities in large scale open source projects. #AI4code Read more at:

Software vulnerabilities can be notoriously time-consuming for developers to find and fix. Today, we’re sharing details about CodeMender: our new AI agent that uses Gemini Deep Think to automatically patch critical software vulnerabilities. 🧵

I’m stunned by the podcast with @dwarkesh_sp and Richard Sutton. He wins a Turing award for his contributions on AI and comes on and doesn’t even understand the basics of modern LLMs. He’s like another Gary Marcus. It was very sad to see Dwarkesh trying to be respectful and

ثورة في تقديم نماذج الذكاء الاصطناعي! 🚀 Databricks تكشف عن تقنية GEPA التي تعد بتوجيهات أذكى وتخفيض التكاليف حتى 90 مرة، مع وصول أصيل إلى GPT-5 بفضل الشراكة مع OpenAI. اكتشف مستقبل المؤسسات الذكية الآن. https://t.co/HiM3c1pOar #الذكاء_الاصطناعي #Databricks #GPT5 #تقنية

Code World Model from FAIR isn’t the first model trained by interleaving interpreter state with code tokens. The first paper was the “Scratchpad” paper in Nov 2021, which was also cited by @_jasonwei’s CoT paper. https://t.co/LC5jpYOLgi

Vibe construction

If you do research related to code translation, transformation, and modernization, consider submitting your research to ReCode workshop and attend it (colocated with @ICSEconf in Brazil)! Check these tweets and workshop website for more details.

Very excited to see our Gemini models getting better and better at coding! An advanced version of Gemini 2.5 Deep Think at the 2025 International Collegiate Programming Contest (ICPC) World Finals achieved gold-medal level performance! 🎉 https://t.co/yCQKOagnkm