Yoav Goldberg: If you can write a Python program for a task, it's not worth RL-ing over
yoavgo · x · 2026-11-03
A debate over whether a model performs on MinAtar games "given the rules and code" drew a crisp take from Yoav Goldberg (yoavgo): if you can write a Python program to perform the task, it's not worth RL-ing over.
Thread context:
- MarlosCMachado challenged that handing the model the MinAtar rules and code makes it an altogether different research question;
- twkillian conceded the point — especially when game mechanics and rewards are fully visible or searchable — saying he was never impressed, just fascinated by the hackability of these problems;
- Goldberg's reply distills the RL boundary question: tasks expressible as explicit rules shouldn't be solved with RL.
More from AGI Musings
- Anthropic roadmap: AI could fully automate top research teams by early 2027 — maksym_andr · 2026-11-03
- Vibecoding discourse is hypnotizing normal people into paying for tools they don't need — arthurcolle · 2026-11-03
- The real AI threat to SaaS: AI-first small teams undercutting Slack at a fraction of the price — bendee983 · 2026-11-03
- AI adds new office chores, and women are most likely stuck fixing the slop — AnnaCiaunica · 2026-11-03
- Claude's 10 tips for the next decade of AI: run open models, avoid vendor lock-in — alex_verem · 2026-11-03
- Scoble shares on-demand CNC printing with agents making parts overnight — Scobleizer · 2026-11-03