LLM/VLM-based OCR is highly prone to hallucination - the model *does not know* w... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		rafram 81 days ago \| parent \| context \| favorite \| on: AI is going great for the blind (2023) LLM/VLM-based OCR is highly prone to hallucination - the model does not know when it can’t read a text, it can’t estimate its own confidence, and it deals with fuzzy/unclear texts by simply making things up. I would be very nervous using it for anything critical.

paulsutter 81 days ago [–]

There are really amazing products coming

rafram 81 days ago | [–]

I’ll believe it when I see it.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact