I’m working on a theatre recommendation engine for personal use, scraping listings from theatres, and automatically looking up and gauging reviews from a review aggregator.
If I crtl+A such a review page and drop it into a LLM with the prompt to create a short summary of all reviews and score them, the result is close to perfect immediately. But using the API to GPT4o doesn’t work nearly as well despite several attempts at improving the prompting (the review page has been fetched and parsed into a JSON file).
Instructing the LLM to “edit this script to produce summaries like in the perfect example I’ve given you” yields optimistic statements, but not the desired results.
Should I try a different LLM? Otherwise it appears I should drop the API idea and just try and automate copy pasting the entire review page into the web interface?