3 ms·

Multimodal LLMs are not up for these tasks imo. It can describe an image but its not great on tables and numbers. Now on the other hand, using something like Te

Multimodal LLMs are not up for these tasks imo. It can describe an image but its not great on tables and numbers. Now on the other hand, using something like Textract to get the text representation of the table and then feeding that into a LLM was a massive success for us.

is_true 2y ago
LLMs don't offer much value for our use case, almost all values are just numbers
- infecto 2y ago
  Then you should be using something like Textract or other tooling in that space. Multimodal LLMs are no replacement.
  - is_true 2y ago
    We use opencv + tesseract and easyocr