jessicata comments on Foom & Doom 2: Technical alignment is hard

[jessicata]

Relevant paper: Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?

YouTube explanatory video