🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

This papers 2.0 could be teacher student self-play arxiv.org/pdf/2506.08388

arxiv.org/abs/2505.22954

👀 Exploratorium

What if an LLM could update its own weights? Meet SEAL🦭: a framework where LLMs generate their own training data (self-edits) to update their weights in response to new inputs. Self-editing is learned via RL, using the updated model’s downstream performance as reward.

a poem a day keeps the blues away I’ve been collecting the poems I read in this Google sheets spreadsheet site. Each emoji opens to a new poem link. This map has been handcrafted for the past 2 years on and off, and I hope you will share your favorite poems -> 💌

kind of a surreal moment being on stage with Lisa Su as she announces & thanks us for the competition we built the past year building w/ Matej Sirovatka Mark Saroufim, Ben, & Erik (all in our free time :p) on GPU MODE has been genuinely incredible, can’t thank you guys enough ❤️

Very proud to announce Cohere’s partnerships with Canada and Britain! Thank you Prime Ministers Carney and Starmer for your vision and leadership. 🇨🇦 ❤️ 🇬🇧

Huge congratulations to Anna Goldie on receiving her Stanford University PhD today! It’s been a great journey!

Cartridges, powered by Tokasaurus! 🤝⚡️🦖

Still feels surreal to have been on stage with one of the greatest CEO's of our time Dr Lisa Su where she explicitly called out GPU MODE and the work we did to enable the world's first $100K competitive kernel competition I never in a thousand years would have imagined that a

Introducing ALE-Bench, ALE-Agent! Towards Automating Long-Horizon Algorithm Engineering for Hard Optimization Problems Blog: sakana.ai/ale-bench/ Paper: arxiv.org/abs/2506.09050 ALE-Bench is a coding benchmark primarily focused on hard optimization (NP-hard) problems. We

"type with the fury of a thousand devins"

welcome to the RL era. anything that can be simulated, will be solved.

And we made t-shirts for everyone who finished the course: x.com/simonguozirui/… If you see someone wearing this shirt, watch out - they're dangerous. Congrats to everyone who put in the hard work and finished!

So excited to be having this session with Johannes Hagemann and Jackmin from Prime Intellect! Now more than ever, efficient verifiable inference methods are critical to powering decentralized and trustless AI services. Come by to dive deeper into their novel approach :)

Here's how Gemini 2.5 Flash-Lite writes the code for a UI and its contents based solely on the context of what appears in the previous screen - all in the time it takes to click a button. 💻 ↓

the model keeps on training even when the underlying infra keeps failing....out-of-the-box PyTorch