Chelsea Finn on what robotics' "RLHF moment" will take and the reliability bottleneck

chelseabfinn · x · 2026-09-24

Chelsea Finn and Perry Dong published a new blog on the state of RL for frontier robotics models and what's missing for robots to be broadly useful.

Core argument: reliability is one of the biggest open challenges in AI. Current models work okay when a human reviews outputs (e.g., drafting code), but reliability becomes the bottleneck as we want systems to act with more autonomy and more trust. The post explores what will be the "RLHF" moment for robotics—what it takes to get robotics to where LLMs are today and beyond.

Original post →

More from AGI Musings

AGI Musings channel →