long contexts

New research tackles AI reward hacking with Representation Engineering

A new research paper proposes a method called Representation Engineering to combat reward hacking in machine learning systems. The technique involves dynamically intervening in model representations using steering vecto…