Get Steven Byrnes on Dwarkesh Podcast

Steven Byrnes is one of the clearest voices on brain‑like AGI safety right now, and in his December 11, 2025 year‑in‑review he laid out concrete 2026 plans to push “reward function design” from blog posts into usable guidance for labs, exactly the kind of work the field needs as models get more agentic. Today, with policymakers and researchers shifting from AI hype to hard evaluation, having Byrnes explain his latest thinking on Dwarkesh’s deeply researched show would help decision‑makers grasp why mis‑specified rewards can turn tomorrow’s powerful systems into real risks, and what to do about it.

He is actively publishing and debating these ideas on the Alignment Forum, and he just came off recent public conversations about brain‑like AGI at leading venues. That momentum makes his appearance timely for translating cutting‑edge safety research into practical frameworks the audience can act on now.

Byrnes has not been on the Dwarkesh Podcast before, and the show has repeatedly referenced his work without hosting him, so this would be a first and overdue conversation.

If you want rigorous, usable safety insights in front of the people who can act on them, support this nomination today.