Causal Concept Graphs in LLM Latent Space for Stepwise Reasoning
arXiv:2603.10377v1 Announce Type: new Abstract: Sparse autoencoders can localize where concepts live in language models, but not how they interact during multi-step reasoning. We propose …
Md Muntaqim Meherab, Noor Islam S. Mohammad, Faiza Feroz
9 views
