Fetching the paper…

AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders · Around