I've been trying to use ChatGPT for my work in Union, HR, and Law, mainly to support workers and improve their working conditions. I often refer to PDFs of the Union contract and Worker Representation Law, which are all publicly available. My goal is to use the most up-to-date versions of these documents to answer questions like whether a new work model proposed by management is allowed, and what loopholes I should be cautious about. I want it to provide arguments for discussions with management, so I even created a custom GPT utilizing these documents.
However, I've been facing issues for several months now. ChatGPT started giving me incorrect answers, which diminished my confidence in using it. I briefly switched to Gemini 2.5 flash, which provided better answers, but now it seems to be less effective. I've tried simplifying the Gem-Description but that didn't help much.
I've heard that some models (like o3) can't interpret PDFs as well as others (like 4o). Should I switch back to a different model or maybe consider using another LLM that would be better for my specific needs?
3 Answers
Honestly, it sounds like these models don't enjoy the complexities of your work! If it gets too tricky, maybe exploring a more specialized tool would be worthwhile.
These models can be a bit tricky and often need clear guidance to provide accurate information. If you had better results with 4o, I’d recommend going back to that version. Have you also considered using Google NotebookLM for analyzing PDFs? It can be really helpful for your needs.
Great suggestion! I’ve heard good things about NotebookLM for PDF use cases. Glad to hear it worked well for you!
Related Questions
Neural Network Simulation Tool
xAI Grok Token Calculator
DeepSeek Token Calculator
Google Gemini Token Calculator
Meta LLaMA Token Calculator
OpenAI Token Calculator