Factorizing a model's final unembedding into a sparse feature basis for logit-lens analysis.
🤝 Open to Collab
Matteo He
hematteo
AI & ML interests
Generative AI, Mechanistic Interpretability, AI Safety, Post-Training, Reinforcement Learning
Recent Activity
updated a Space 1 day ago
hematteo/readout-feature-atlas upvoted a paper 2 days ago
Self-Evolving Agent Harness liked a model 2 days ago
meta-models/Muse-Spark-Open-816B-A42B