
[AIEWF Preview] Multi-Turn RL for Multi-Hour Agents — with Will Brown, Prime Intellect
Latent Space: The AI Engineer Podcast
00:00
Navigating Taste in Multi-Agent RL
This chapter explores the concept of 'taste' in research among graduate students, particularly within multi-agent reinforcement learning (RL). The speakers discuss their experiences in addressing overlooked questions, the challenges of encouraging AI models to use tools effectively, and the complexities involved in verifying model performance. They highlight innovative strategies for enhancing reward systems and the importance of flexible approaches in upcoming RL projects.
Transcript
Play full episode