AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders | Read Paper on Bytez