DigiNews

Tech Watch by Johan Denoyer

← Back to articles

Many SWE-bench-Passing PRs Would Not Be Merged into Main

Quality: 8/10 Relevance: 9/10

Summary

A METR study compares SWE-bench automated grading vs maintainer review for AI-generated patches. It finds a ~24-point gap and cautions against over-reliance on benchmarks; discusses robustness checks and time horizon findings.

🚀 Service construit par Johan Denoyer