I have been finding AI really useful for coding, but this is a good reminder that AI benchmarks are not representative of reality, and developers cannot be trusted to accurately estimate their work, even after completing it 🙃
https://bsky.app/profile/metr.org/post/3ltn3t3amms2x
