Rishi Bommasani
What language-model developers are and aren't telling us about their own models — and how to measure that systematically.
Most public LLM evaluation measures what a model can do. Bommasani's Foundation Model Transparency Index, run out of Stanford CRFM, runs along a perpendicular axis: how much we know about the model itself — what was in its training data, which internal evaluations were performed, which downstream impacts have been documented, what has been disclosed publicly versus kept private. The underlying argument is direct: capability benchmarks mean little when the conditions that produced those capabilities are not on the record, and treating transparency as a separate research discipline (rather than a PR concern) is the only way the comparison gets honest.