RESEARCHERS SAY GENERATIVE AI ISN'T REPLACING DEVS ANY TIME SOON

Summary from the TLDR Web Dev newsletter

A study conducted by AI researchers from Princeton and the University of Chicago suggests that LLMs are far from being capable of solving common software engineering problems. The study evaluated the performance of various LLMs on over 2,300 real GitHub issues and found that they could only solve such problems correctly in a very small percentage of cases, with the most advanced models achieving only 4.8% success. LLMs often struggle with the complexity and context of real-world software engineering tasks.