Gary Marcus Assesses LLM-Generated Code Patches

Gary Marcus stated that the expected value of a fully LLM-generated, non-human-reviewed patch is a net-negative by a considerable margin. The cognitive scientist and author, known for critiquing deep learning, made the comment in a public post. It directly addresses the use of large language models to produce software patches without human oversight. The remark stands as his assessment alone, with the source line presenting it as a quoted statement from him.