World Wires · story 31280 · corroborated · 1 source(s)

Predicting Alignment Generalization with Value Representations

This paper examines how narrow post-training behaviors in LLMs influence generalization across unseen contexts and environments.

Open in the desk

Coverage

• Predicting Alignment Generalization with Value Representations arXiv: new preprints (security, AI, physics) · tier C · clear net · 2026-10-08 17:47 UTC

What this site indexes