OpenAI says it found widespread task issues in SWE-Bench Pro, estimates ~30% of tasks are broken, and retracts its earlier recommendation to adopt the benchmark (OpenAI)
2026-07-09
·
via Techmeme
OpenAI : OpenAI says it found widespread task issues in SWE-Bench Pro, estimates ~30% of tasks are broken, an…
此内容由惯性聚合(RSS阅读器)自动聚合整理,仅供阅读参考。 原文来自 — 版权归原作者所有。