@betterstack: An LLM in Just 200 Lines of Python?! (microGPT) #gpt #ai #llm

Better Stack
Better Stack
Open In TikTok:
Region: CZ
Friday 20 February 2026 09:22:08 GMT
102210
4617
35
837

Music

Download

Comments

shyn_ca
Shyn.ca :
I'm terrified of AI. it's just a glorified copy and paste. But people think it's thinking
2026-03-05 10:28:54
15
epistatesinc
Epistates :
It works great, I'll be posting a video soon
2026-03-29 20:26:40
1
mrdonmomo
mrDonMomo :
The real flex here isn't the 200 lines, it's that people finally see the math behind the magic isn't rocket science. microGPT is basically the 'hello world' of understanding transformers. If you get this, you get 90% of what ChatGPT does under the hood.
2026-03-03 02:26:38
25
concentratedcake
Concentrated Cake :
You know what the chainrule and autograd are but not the adam optimizer?😅
2026-02-20 18:39:58
30
kakayenu_shujaa
Kaka Yenu Shujaa :
Karpathy is a Kendrick Lamar of A.I
2026-02-28 15:46:41
6
blaised23
dablaise :
He’s a Fking legend
2026-03-03 01:49:16
1
daily.with.missy
Missy :
Hahaha this is so
2026-05-18 23:20:02
1
mjdale2003
mjdale2003 :
It’s what us engineers usede to have to do in our head😂
2026-03-09 02:57:08
1
m.salah2405
Mohamed Salah :
It’s worth giving a shot
2026-03-05 06:39:20
0
alexandra.build.o
Alexandra.build.ops :
Oh, want to try!
2026-02-21 08:25:38
1
michaelearle75
michael :
well - I was looking for just this 😳 thank you 😊 👏👏👏👏👏👏-
2026-02-21 04:46:34
0
krakecz3
Krake :
gg thanks for info have idea 😂💪
2026-02-20 16:00:32
1
mobilenorank
Mechanical Electrical Store :
still halusination😂😂😂
2026-03-13 18:55:40
0
finalspy
FiNaLsPY :
lack a bit of efficiency as in python ... processes 5 docs/s mono cpu => pass it in rust, add multi cpu and you go to 5000 docs/s ... see https://github.com/finalspy/MicroGPT yet not gpu efficient, needs full rewrite 😏 happy experiment
2026-02-22 11:33:00
5
moomoo4123
moomoo :
who can see it's face in the QR code image?
2026-02-23 00:29:05
1
nita59335
Nita :
Scanning stuff is just another way to get hacked
2026-02-21 11:18:33
3
gerald_alphaxad
Gerald Alphaxad :
That tiny code will eat your hardware for lunch
2026-03-18 07:07:25
0
cytus101
Tester 𓆩⟡𓆪 :
😁😁😁
2026-05-20 09:27:52
0
t3d.uk
Ted :
🔥🔥🔥
2026-02-20 10:22:15
0
automated.sketch8
AutomatedAIEngine :
whole Idea in one line: This code teaches a computer read simple text data turn letters into numbers guess the next letter check how wrong it was adjust itself a little repeat many times then generate new text
2026-03-07 05:35:18
0
alseenwalnoot
walnoot :
Finally someone that optimizes code instead of “optimizing” then making it unreadable and depend on a specific python version from 2019 and 200 libraries
2026-02-20 22:17:51
15
sriious
SRIIOUS :
is literally the algorithm that AI models use so, it's only limited by the hardware capacity you provide to it... long story short...
2026-02-21 18:15:16
8
dmn613
Dan Neuman 🇨🇦💚🇪🇺🇲🇽 :
Sounds like it's pretty slow, since there are no C modules used, other than `math` and `random`. Someone rewrote it in Typescript and claimed it was 8x faster! I think if you adapted this to Cython (compiled Python) you could get to about 80% of pure C speeds.
2026-02-21 06:14:44
2
nothanks948
Troy southbound :
It’s amazing… but I haven’t actually used it yet.
2026-02-22 09:49:07
0
mariomaletic
Марио Малетић :
Omg
2026-03-06 07:27:07
0
bolsilloajeno
wlock :
what's that guy's GitHub profile? everyone sharing this guy's work on gb too
2026-03-09 04:04:25
0
To see more videos from user @betterstack, please go to the Tikwm homepage.

Other Videos


About