00:18
2026-10-16
provenbrief.com
ai-safety
The AI benchmark dispute ledger: every fully sourced case to date of a model gaming, disputing or correcting its score
Only one of the six fully sourced entries in a new benchmark dispute ledger ended with a corrected score: IQuest-Coder-V1's 81.4% claim on SWE-bench was revised to 76.2% after researchers found 24.4% …