Post #480672

I learned an interesting thing while working with an AI agent to help me transcribe portions of a terribly formatted pdf so it could be transcribed. I can't go into details on what it was, but I was way more successful asking it to extract the document page by page, sometimes paragraph by paragraph. Apparently, because large language models look for patterns in documents like these, you're literally going to get an approximation of the document. Without processing the document piece by piece, I got a literal hallucination of a document including topics that were similar to the doc, but they were hallucinated patterns rather than the exact text. It was a lot of work, but I was able to transcribe mathematical info including descriptions of graphs and it turned out well. Exactly why collaboration with an littlem is much more effective than expecting it just to spit something out.