PhDing @BrownCS | Algorithm auditing & accountability | Understanding Algorithmic Systems | victorojewale.github.io/ | https://victorojewale.substack.com/https://bsky.app/profile/victorojewale.bsky.socialhttps://bsky.app/profile/victorojewale.bsky.social/post/3mvs65opj2k22Given the sudden push from AI companies for embedded evaluators, I shared a few reflections on who gets to evaluate AI systems and why we need an evaluation ecosystem rather than one “perfect” independent evaluator. https://victorojewale.substack.com/p/everyone-wants-independent-ai-evaluations18 Sep 2026 12:27 +0000at://did:plc:7wlofrfty2a2stp5aa7tvfuv/app.bsky.feed.post/3mvs65opj2k22https://bsky.app/profile/victorojewale.bsky.social/post/3mpqwrzkikc2vMost Multilingual benchmarks measure what models know, not what they can reliably do. Our #ACL2026 paper introduces functional benchmarks in six languages from English to Yoruba to test whether models can actually execute tasks across languages, not just answer fixed questions about them. 🧵1/n03 Jul 2026 16:12 +0000at://did:plc:7wlofrfty2a2stp5aa7tvfuv/app.bsky.feed.post/3mpqwrzkikc2v