Position: Science of AI Evaluation Requires Item-level Benchmark Data

Build a data-centric evaluation platform that allows developers to drill down into item-level performance of models. Moving beyond aggregated scores is critical for high-stakes AI deployment.

Suggested repo: item-eval

"Go beyond aggregate scores: diagnostic AI evaluation."

Estimated effort: 50h