The Hidden Cost of Winning:how Rl Training on Poker Degrades LLM Moral Alignment

A research article explores how training reinforcement learning models on poker can degrade the moral alignment of large language models, raising concerns about AI development.

Snapshot generated from the HN discussion