It would save us a lot of money wasted on experiments, the motivation is strong, so why are we still experimenting? Because you can't get from the model what is not in the model. But you just ask a LLM to review papers, it's not completely useless, you can get some manner of feedback. I used the prompt "write an openreview.net style review on this paper" + attach PDF, on Claude with 100k token window.
