100 million context window？ Cant wait to try‼️

Is OpenAI's o1 a good calculator? We tested it on up to 20x20 multiplication—o1 solves up to 9x9 multiplication with decent accuracy, while gpt-4o struggles beyond 4x4. For context, this task is solvable by a small LM using implicit CoT with stepwise internalization. 1/4

Very proud to introduce two of our recent long-context works: HELMET (best long-context benchmark imo): shorturl.at/JnBHD ProLong (a cont’d training & SFT recipe + a SoTA 512K 8B model): shorturl.at/XQV7a Here is a story of how we arrived there

And a few papers in finding/demo/industry tracks. Come find us! A lot of us are here including Ansong Ni Sophia S. Han Yixin Liu Yilun Zhao Alan Li Linyong Nan Charlie Gao

New paper out! Happy to collaborate with Junjie!

Alibaba's CoCreate 2025 just launched in Las Vegas --I'm thrilled to see our AI sourcing agent Accio in the spotlight. 🚀 Proud to work with Alibaba for the past few months, leading the design and integration of reinforcement learning to enhance agents for product DeepSearch. 🔧

Nathan Lambert

How often do you see two professors (from CMU) to jointly present their poster? Graham Neubig Daniel Fried A must-check! 😆😆😆

Agents are joining us at work -- coding, writing, design. But how do they actually work, especially compared to humans? Their workflows tell a different story: They code everything, slow down human flows, and deliver low-quality work fast. Yet when teamed with humans, they shine

Excited to see growing attention on the challenge of information aggregation! Our ACL paper also tackles this in long-context understanding with aggregate reference-based information. Check it out: wujunjie1998.github.io/Ref-Long-websi… Frank Gu Arman Cohan

刚才想到，LLM 的确没派生出智慧，但是它发明了一种方法让人们共享智慧，就像最近三十年我们在互联网上共享信息一样。

me stepping down. bye my beloved qwen.

working on distillation

We're releasing a technical report describing how Composer 2 was trained.

another jeff dean jokes haha

crazy

wrote a guide on getting compute grants as a student, something I wish I did more at the beginning of my PhD. It's honestly one of the highest ROI things you can do as a student (we've gotten 100k+ gpu hrs for roughly 2 weeks of work writing). nightingal3.github.io/blog/2026/04/1…

今天非常高兴和大家正式介绍并开源我的第 14 款 vibe 产品 wanman.ai 它的理念很简单，让世界上所有人，都能在 AI agents 团队的帮助下，从零创办或接管任何组织，围绕用户的核心意图，持续自动化地运营一人公司。 为了实践这种理念，wanman 必须设计的尽可能简单，不需部署，不用买