A new analysis from OpenAI reveals issues in SWE-Bench Pro, a popular coding benchmark, raising concerns about reliability and accuracy in evaluating AI models.
Origine de l’article : lire l’article original
Separating signal from noise in coding evaluations
