Github Signals
Region: CA
Sunday 05 July 2026 04:40:00 GMT
52638
1941
46
273
Music
Download
Comments
bob :
the fact something as stupid as this is needed doesnt really give confidence in the future of llm and ai
2026-07-06 17:32:47
53
square caboose :
LLMs cannot "read" images. They use a smaller model to produce a read-able text format for the real model to use. Once the text format has been created, you're not saving tokens anymore. This helps a lot with short tasks, but agentic work it will maybe cut costs by 1% tops.
2026-07-05 18:25:08
32
acidayan :
will i get the same output as text?
2026-07-05 16:11:23
5
Wodoame :
AI cost is bringing out the ingenuity in us 🤣.
2026-07-05 11:12:19
35
Six :
i use math instead it can save up to 50%, still doing tests and optimizing
2026-07-09 22:08:09
1
waningus :
But i tried it once and ask chatgpt to read a big text inside a big image and it failed and with more tries it read some stuff wrong
2026-07-05 14:04:14
8
junww6 :
The model cannot accurately see that. I mean it is entirely possible that you end up with a wrong instruction read, which is worse than no instruction
2026-07-05 16:09:47
1
MonoluminAI :
U sure that it works?… usually the images is not so light 😳
2026-07-08 09:07:32
0
Jesse¹⁰²⁵ :
cocoindex_code with a small embedder 70% savingon code base
2026-07-08 04:23:42
0
user4764533001770 :
Proxy as a MITM, but selling to you as a benefit, is a good way to get data out of a close loop setup
2026-07-06 00:10:23
1
Zuck :
ridiculous
2026-07-06 20:34:14
0
voidapp.lol :
Do i look like i case chlling on my 3 claude max plans?
2026-07-07 07:54:43
0
Archmagos X :
Yea its limited to OCR
2026-07-05 14:33:54
6
lok1s :
gemini when the text on the image is too small starts hallucinating hard, i think Claude won't be that much better
2026-07-05 15:55:06
4
itsyoboi.van :
Input tokens is hella cheap, why bother and with this agent harness, planning is better than long prompts.
2026-07-05 22:51:50
4
AJ P :
you always say that then increases those already existing
2026-07-05 12:44:43
0
🩸 :
This doesnt reduce costs, just use smaller models for the prompts
2026-07-05 20:53:37
0
SERGIO :
definitely trying this out😜 sorry ai overlords
2026-07-06 15:38:16
0
Unemployed :
Performance would be worse
2026-07-05 17:55:39
0
Mattias Aabmets :
gonna get patched in like 3 days
2026-07-06 20:04:02
0
To see more videos from user @github.signals, please go to the Tikwm
homepage.
