EVOLVING REPRESENTATIONS IN LANGUAGE MODELS

Summary from the TLDR AI newsletter

Google researchers reveal how internal model representations shift over conversations, showing that concepts like factuality are fluid and context-dependent, with implications for model steering and interpretability.