Spring 2026 course team expands Inspect AI notebooks

A project team has revised and expanded four Jupyter notebooks that provide a practical introduction to building AI evaluations with Inspect AI. The tutorials guide users from basic evaluations to benchmarks, LLM-as-a-judge methods, and autonomous agents, while also updating the official Inspect AI tutorial with a new custom solvers section. The authors argue that safety evaluations are unusually critical because AI safety lacks the rich feedback signals available during capability training. Open article →