lib/oppress

LLMs structure moral knowledge geometrically, research finds

A new research paper explores how large language models (LLMs) organize moral knowledge, moving beyond simple moral content detection. The study found that LLMs distinguish between different moral foundations, such as c…