🚀 Come chat with @_chris_lu_ who is presenting his poster on discovering Preference Optimisation Algorithms for LLMs, using LLMs.
Happening now @AutoRL_Workshop, Stolz 0
Foerster Lab for AI Research
95 posts
Foerster Lab for AI Research
@FLAIR_Ox
ML research group @uniofoxford. Focussed on multi-agent, open-ended, meta and reinforcement learning as well as agent based models. More at foersterlab.com.
Joined March 2022
- 👀 Language models trained to play Among Us with MARL!Amogus (AAMAS 2025)
00:00 - 🤖 @AlexDGoldie is presenting his work on Learned Optimisation Algorithms for RL. Work done with @_chris_lu_ and @JacksonMattT. Come check it out! Happening now, Stolz 0
- 1/🚀 FLAIR is coming to #icml2024 in Vienna 🎉 we are very excited to share our work with you! You can find us here ⬇️✨ Or use shorturl.at/Qw8QN 🔗 for clickable links
- 🔥 OPEN shows that learned optimisation produces huge performance improvements in RL, and can generalise zero-shot (without any hyperparameter tuning)!🇨🇦 I'll be at #NeurIPS2024 next week to present our spotlight paper on learned optimisation for RL! If you're interested in (meta-)RL or the potential of learned algorithms for ML, I'd love to chat! 📨 DM if you'd like to get a coffee!
- ⚒️ @mitrma is presenting his #ICML2024 spotlight, Craftax, with @benjamin_ellis3 and @mcbeukman at poster 1310! Come learn about the power of JAX and Open-Endedness, or try your own hand at the game to see if you can beat an agent’s score!
- 📈 @scowardai is also presenting his work on Higher-Order and Self-Referential Evolution which improves robustness to evolutionary hyperparameters. Work done with @_chris_lu_ Come check it out now! @AutoRL_Workshop, Stolz 0
- Sampling for learnability demonstrates significant benefit in RL autocurricula! Come check out its poster at #NeurIPS2024!Hello! I'll be at NeurIPS next week presenting our work on using learnability to select levels for RL autocurricula. If you're there, I would love to chat about curricula and RL generalisation more broadly. Please DM if you'd like to grab a coffee :)
- 🧑🔬 FLAIR is presenting three more great papers today at #NeurIPS2024! Come talk to us and find out what we've been doing!
- 🔓Unlocked: train a generalist RL agent on a single GPUWe are very excited to announce Kinetix: an open-ended universe of physics-based tasks for RL! We use Kinetix to train a general agent on millions of randomly generated physics problems and show that this agent generalises to unseen handmade environments. 1/🧵
00:00 - M-FOS provides a simple and powerful approach towards opponent-shaping (i.e. learning in a meta-game) which overcomes the key three limitations of prior work: myopic shaping, higher order derivatives and symmetry breaking. We also show that powerful shaping can lead to extortion.General-sum games describe many scenarios, from negotiations to autonomous driving. How should an AI act in the presence of other learning agents? Our @icmlconf 2022 paper, “Model-Free Opponent Shaping”(M-FOS) approaches this as a meta-game. @_chris_lu_ @TimonWilli @casdewitt 🧵
00:00 - The AI Scientist has the potential to transform science by completely automating the research pipeline: from ideation, to experimentation and execution! Incredible work led by @_chris_lu_, @cong_ml and @RobertTLange!Introducing The AI Scientist: The world’s first AI system for automating scientific research and open-ended discovery! sakana.ai/ai-scientist/ From ideation, writing code, running experiments and summarizing results, to writing entire papers and conducting peer-review, The AI
- Super greatful that @SakanaAILabs has helped advertise our joined work while we were locked out of the account due to a bug on @X which is now fixed. Well, we are back now and the work is still amazing so no harm in reposting!Can LLMs invent better ways to train LLMs? At Sakana AI, we’re pioneering AI-driven methods to automate AI research and discovery. We’re excited to release DiscoPOP: a new SOTA preference optimization algorithm that was discovered and written by an LLM! sakana.ai/llm-squared/
00:00 - Replying to @FLAIR_Ox😈 @silviasapora is also presenting her work, EvIL, about using evolution to aid generalisation in imitation learning! Super cool - poster 1208! Check it out!!
