A new dataset called GoodPoint-ICLR contains 19,000 papers and reviewer comments annotated via author responses. Researchers used these success signals to train a model that generates more actionable, valid feedback for scientists. This approach shifts LLM utility from automating research to augmenting the peer-review process. It provides a concrete framework for improving academic presentation.