this post was submitted on 21 Oct 2025
15 points (85.7% liked)

Futurology

3372 readers
49 users here now

founded 2 years ago
MODERATORS
voidx@futurology.today
Lugh@futurology.today
Espiritdescali@futurology.today
AwesomeLowlander@futurology.today
15
DeepSeek has open-sourced DeepSeek-OCR, a 3B-parameter model that upends how AI will deal with text, by being far more efficient than anything else available. (eu.36kr.com)
submitted 6 days ago by Lugh@futurology.today to c/futurology@futurology.today
3 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[โ€“] doomsdayrs@lemmy.ml 3 points 6 days ago (1 children)

So instead of recognizing characters....

  1. Compress page / text into a handful of pixels.
  2. Feed pixels into a generative AI.
  3. Hope for the best.

I rather just use existing OCRs which can be easily backtracked in how they processed text.

[โ€“] FauxLiving@lemmy.world 3 points 6 days ago

They were able to efficiently encode visual information to be used by further networks. In this case the further network was a language model trained on an OCR task.

The news is the technique, the OCR software is a demonstration of the technique. Encoding visual information efficiently is also key for robotics which use trained networks in their feedback control loops. Being able to process 10 times as much visual data with the same hardware is a very significant increase in capability.