Announcement
@natolambert outlines six open problems in AI post-training
In a new deck, @natolambert links six post-training research questions to personal goals with Trillium Labs.
TLDR
According to @natolambert, the deck covers harness design, model personality, open-source reinforcement-learning environments, reasoning efficiency, and scaling laws for reinforcement learning and multi-agent systems. It largely reflects personal goals with Trillium Labs, not a promise to tackle every project at once. @natolambert also says a video re-recording is coming after a Cornell talk.
Combined views
2.4K
2 Sources, first seen ago
Comments