parthknowsai
Region: US
Monday 27 April 2026 12:09:46 GMT
158768
10219
221
1036
Music
Download
Comments
femfortfive :
this is why we should stop calling Claude and chatGPT AI. They are LLM's. Models. Tools. The wording matters.
2026-04-27 13:03:33
493
jant2534 :
and water is wet
2026-04-27 17:54:56
211
jono :
Without the model the "98%" would be useless and do nothing. LLMs are by definition AI
2026-04-27 22:18:56
61
fuagatik :
and I have to pay 100% of the "tokens" 😂
2026-04-27 12:21:09
170
d0n7b33v1l :
Yes this was already well known. The AI part is just a function that takes text (or tokenized text) and returns a vector of probabilities containing the models entire vocabulary. Then a seperate picker function selects the next word (token) based on probability (with a few tweaks like temperature (randomness). And people are seriously askong questions like "is it conscious?". No clearly its not. Its a function that returns a word at a time.
2026-04-29 18:31:31
3
bidord :
I'm sure more than 98% of YouTube's backend code is not video codecs. Does that mean YouTube is not a video app?
2026-04-28 23:08:08
20
rabibit767 :
That’s a highly misleading claim! You’re essentially saying, if you look at the non-AI component of it, the call to the AI is a small part of that. No sh*t. Is the AI a tiny part of the processing done by it? No, not even close.
2026-04-28 12:01:49
26
deus :
cloud code with deepseek v4 flash works better than opencode + deepseek v4 , i discovered that yesterday experimentally
2026-04-27 12:16:23
26
Healien Rosewood :
They thought AI was a machine for extracting human value. They missed that AI is a mirror-interface for extending human capability.
2026-06-09 11:41:27
1
mike_diamonde :
Infrastructure is as critical as the model itself. Coding IDE benchmarks prove this, the same model and API can yield different results depending on the underlying stack. While the model provides the intelligence, the infrastructure determines the execution. Intelligence is the engine, but infrastructure is the transmission, you need both for peak performance.
2026-04-27 13:10:09
12
LoGyxalPerk :
Well if you build strong ai and strong surrounding code. You get better results than all of them. Ai brains are possible and more powerful, with a good harness around it, you still get amazing results.
2026-05-20 17:45:12
2
Neosilver GK :
What a new discovery 😏 Kind of what normal employee do with the right tools. Nobody writes code on paper anymore. They use linter, formatter and lsp. A good setup saves a lot of time/money.
2026-04-27 12:42:10
27
user8511144463273 :
Uumm not seeing a problem here. I've got a problem, Claude solves it. Need I say more? 😳😳
2026-04-28 21:45:20
1
Otis Carlisle :
Yeah no shit. Who didn’t know this?
2026-04-28 14:38:09
12
fr3klz :
It’s interesting how the philosophy of Claude Code and Codex are so different. Claude Code more harness, Codex minimal harness. Claude is still my DD, but I have to say I think the Codex team is on to something.
2026-04-27 12:40:50
0
emfinal.boss :
It’s like saying Amazon.com is 1 percent backend. No Amazon backend is probably bigger than frontend. But frontend only calls that backend in 1 percent of the frontend codebase.
2026-04-27 12:18:54
13
tomm3hshipp :
the 98% stat makes sense when you realize most of the work is context management and file orchestration, not generation. the ai part is just the reasoning layer on top
2026-04-27 19:13:39
5
Solutions Joe :
😮Yep, Inference is a black box!
2026-04-28 15:56:39
3
WHERE’SMAICOL :
100%
2026-04-27 20:54:09
2
i_am_thoriso :
I thought this was obvious hence the uninformed call them ai and when I explain to my team I say Language Model🤔🤔
2026-04-28 11:28:48
1
JustDaveed :
I thought this was clearly obvious already…🤐
2026-04-27 18:27:19
3
bigBrain :
thanks captain obvious
2026-04-29 14:01:19
1
PisoChat :
What??? Did you expect the tool to be the AI? Of course NOT!!!
2026-04-27 15:25:33
4
🫥 :
What did you expect
2026-04-27 22:23:34
2
maxkan11 :
That’s repetitive tasks optimization. Written by ai
2026-04-28 05:07:07
2
To see more videos from user @parthknowsai, please go to the Tikwm
homepage.
