Fellow Stephen Casper and coauthors propose proportional evaluation approaches to open-weight AI models. They argue that such mechanisms are necessary because existing evaluation practices are able only to accommodate closed-weight models. Read more from RAND.