LLM models are already capable of diagnosing scientific outputs, but, until now, had “no physical agency to actually perform" ...