Hi there,
This is my first post and it’s about potential failures when reading html/pdf files (ChatGPT-O) - oftentimes content is generated with pictures clearly mixed with text (non-standard icons, diagrams in technical pdfs).
This produces false understanding by ChatGPT and consequently the wrong answer.
However! If the same content is uploaded as a picture, it is much better.
HTML pictures-inside-text are easily detected.
Diagams and their importance maybe less so - but if there is a command ‘read as graphic’, then it will be solved.
There. That was my tried and retried and proven improvement. But please, feel free to comment and add opinion and present solution.
Sincerely,
Mouth of Sauron