🔥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

Recordings of the talk are now available on YouTube: youtu.be/eKOwqay73pA?si…

SWEET-RL: Training Multi-Turn LLM Agents on Collaborative Reasoning Tasks New paper from Meta introduces a new multi-turn LLM agent benchmark and a novel RL algorithm for training multi-turn LLM agents with effective credit assignment over the multiple turns.

Sweet! 🍭 New paper about training Multi-step Agent LLM. If a DPO-based critic has extra information during training, it can train a better LLM agent

Got our first "obviously" LLM generated review. I should have known when we were working on LM ten years ago that it will come back and bite us 😂. But seriously, reviewing feels broken beyond repair.

Attention operates at token-level, but sometimes what we’re looking for have multiple tokens. MTA makes it possible to condition attention on multiple tokens. Super fun work!

We have been cooking! 👨‍🍳 🧵(1/6)

🚨This week's top AI/ML research papers: - Inference-Time Scaling for Generalist Reward Modeling - Multi-Token Attention - Why do LLMs attend to the first token? - Command A - LLMs Pass the Turing Test - Advances and Challenges in Foundation Agents - PaperBench - Effectively

4 advanced attention mechanisms you should know: • Slim attention — 8× less memory, 5× faster generation by storing only K from KV pairs and recomputing V. • XAttention — 13.5× speedup on long sequences via "looking" at the sum of values along diagonal lines in the attention

Back in the day FAIR - 10 years ago pics! When Sainaa worked on Memory Networks, pre-Transformer (see RT).

Nice to see our Coconut paper got featured

Google friends & ex-colleagues -- Google scholar seems pretty broken😔. Our most cited paper from last year "Self-Rewarding LLMs" has disappeared! Scholar has clustered it with another paper (SPIN) and it isn't in the search results. This is bad for PhD student & first author

Really excited to give a talk here after 10 years 🎉 RAM workshop is about "Reasoning, Attention, Memory" and those topics had huge impacts in the last decade of AI. So there would be plenty to reflect and look forward to!