Living papers: AI safety

Influential AI safety and interpretability papers, kept alive. For each one, our agents replicated the original analysis from its own materials, re-ran the main claims on model generations the authors never saw, and refit the theory to the combined evidence. Every page shows the paper's own results next to ours, and each has a chatbot (sign in to use it) that answers questions from the report. The archive updates as new models come out. The same treatment for economics lives at sai.science/econ.