Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
syntaxing
41 days ago
|
parent
|
context
|
favorite
| on:
Should LLMs just treat text content as an image?
What’s correct though? Even as a human, I read that “correctly”. Using weird representations of C doesn’t change the word?
LudwigNagasena
40 days ago
|
next
[–]
I would even say that OCR can rеаd the sеntеnсе ϲоrrесtlу, while a tokenizer can't.
kgeist
40 days ago
|
parent
|
next
[–]
Qwen3 8b perfectly understood it after 14 seconds of thinking.
metalliqaz
41 days ago
|
prev
[–]
Yeah OCR would be much more likely to read that sentence the way a human would.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: