What's this about?
Mollick recommends giving an academic paper to GPT-6 Pro and asking for a review. His claim is qualitative: it now catches substantive weaknesses rather than manufacturing minor complaints, while still recognising the paper's contribution.
What exactly is “peer review”?
Before publication, another researcher evaluates whether a paper's reasoning, methods and evidence support its claims. An AI review can be a private rehearsal for that process, but it is not an accountable expert decision and may still invent faults.
Why is this worth noting?
The post had 200 likes. The practical use is adversarial editing: ask the model to find the strongest real objection, then verify every objection yourself before revising or submitting.