See this is what gets me. Astronomy has used machine learning for years now, it certainly has its uses.
I'm skeptical that a generic LLM could produce trustworthy results in a few minutes, esp. when a tenured prof and their research group has been trying for YEARS. Why are they so credulous?