0
I uploaded the same report to Claude and ChatGPT and asked each for a summary of its main findings. Both responses sounded reasonable, but one emphasized the recommendations while the other focused on the data and caveats.
How much of that difference comes from the models’ interpretation, and how much might come from PDF text extraction or the way the prompt is handled? What’s a fair way to compare summaries when I need to know whether either one missed an important detail?
