00:09
2026-05-27
lesswrong.com
large-language-models
Should we train LLMs to be human?
New research shows that post-training alignment makes large language models less human-like in their responses, raising questions about whether this drift is intentional or optimal. A study introducin…