About half of the AI code solutions that pass the popular SWE-bench benchmark would get rejected by actual project maintainers, according to a new study by the research organization METR.
The article Half of AI-written code that passes industry test would get rejected by real developers, new study finds appeared first on The Decoder.
