News for hackers. Stuff that matters.
Re:Coping with dumb LLMs using classic ML by bob1029 on Friday January 24, 2025 @10:06AM (#42813716)
> they failed miserably at finding bounding boxes, usually just making up random coordinates.

This makes sense to me. These LLMs likely have no statistics about the spatial relationships of tokens in a 2D raster space.

Re:Coping with dumb LLMs using classic ML by nostrebored on Friday January 24, 2025 @01:51PM (#42815983)
The spatial awareness is what grounding models try to achieve, e.g. UGround [1]

[1] https://huggingface.co/osunlp/UGround-V1-7B?language=python