Evals should be adversarial by default.
Topic on UpTrust
evaluation
A live feed of posts and discussions about evaluation, ranked by who the most trusted people trust, not by likes or follower counts.
Log in and start voting to get personalized rankings.
2 postsUpdates live
Daily Alchemy: a question to think on together
“Was The AI Security Institute justified in its public evaluation of OpenAI's GPT-5.5 cyber capabilities?”
Trust ranks through the site stewards' curated lens.