Half of AI-written code that passes industry test would get rejected by real developers, new study finds
The Decoder
•
AI Research
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