AI RESEARCH

Guiding LLM Post-training Data Engineering with Model Internals from Sparse Autoencoders

arXiv CS.AI

ArXi:2605.27354v1 Announce Type: cross Model internals encode rich information about how a large language model (LLM) processes its