Benchmarks as narrative devices by AI builders! Our new paper maps 231 benchmarks highlighted in 139 model releases by 11 model builders in 2025. I'm excited to present “Unsteady Metrics and Benchmarking Cultures of AI Model Builders” next month at #FAccT2026. Paper at arxiv.org/abs/2605.14164

📣 Introducing APAI 2025 — Authenticity & Provenance in the Age of Generative AI Workshop 📅 This October in Honolulu, HI 🇺🇸 📍 In-person @iccv.bsky.social Stay tuned for speaker and challenge announcements! 🔗 sites.google.com/berkeley.edu...

Great work by WITNESS — we need more benchmarks informed by real-world impact like this: arxiv.org/abs/2504.21489