Andrej Karpathy, one of the co-founders of OpenAI, has made almost everything he knows about building AI freely available on GitHub!
From building GPT from scratch to creating AI that autonomously runs experiments, analyzes results, and learns from its own findings.
6 projects with over 300K stars combined ⭐
1. Autoresearch | 92.7K ⭐
An AI that continuously runs experiments, analyzes results, and keeps only the solutions that actually work.
https://t.co/20laxVrgRF…
2. nanoGPT | 64.4K ⭐
Learn how to train a GPT model from scratch. A project that has become the starting point for many people entering the world of LLMs.
https://t.co/1xXnUYLO6U…
3. nanochat | 60.2K ⭐
Build your own personal version of ChatGPT for around $100.
https://t.co/RkAenvkiSp…
4. LLM Wiki | 50K ⭐
A simple but powerful concept for building your own expandable AI knowledge base — something like a second brain.
https://t.co/rN83qAvT7i
5. llm.c | 31.4K ⭐
Train a language model in C, without hiding the technical complexity behind ready-made frameworks.
https://t.co/y6J8Do5gwL
6. micrograd | 18.7K ⭐
Understand how neural networks learn and how gradients work in just a few hundred lines of code.
https://t.co/cdHOKgWdvG…
My recommendation:
Start with micrograd.
Then move on to nanoGPT and nanochat.
This learning path will help you do more than just use AI tools — it will help you understand what actually happens under the hood.
The difference between an AI user and someone who can build intelligent systems starts right here.
Save these 6 repositories. They’re a goldmine of free knowledge.
THIS GIRL, THIS WALL AND THIS BUNNY ARE ALL AI
73,419 likes. 612 comments. 16,305 shares. 19,874 saves
A girl leans in front of a graffiti wall with a taped-up hole. She reaches toward it, a small black bunny pushes its face out, and she bursts out laughing. 13 seconds. Nothing here was filmed. The girl, the wall and the bunny are AI-generated in @Picsart
why it works:
> the caption "it gets crazier" makes you wait for the payoff to the very end
> a dirty wall and grey duct tape make the bunny's clean face the brightest thing in the frame
> her reaction sells the surprise, so you laugh before you think about the render
> 612 comments on 73,419 likes, about one per 120
Did you buy it before the bunny showed up?
SHE PUT HER FEET UP ON THE TALK SHOW DESK AND THE WHOLE STUDIO WENT QUIET
→ A late-night set with a glowing city skyline
→ The host opens with a compliment
→ She answers by lifting both heels onto his desk
→ He can only shrug and call it the best he has seen
None of this exists. The studio, the desk lights, the skyline, two people talking in sync and even the burned-in subtitles are generated, and the only slip is one garbled word in the captions. That is the new bar: a full talk show moment, with timing and reactions, from a single prompt
You can try Seedance 2.5 in
@Picsart
Would you have caught it without the typo in the subtitles?
@0xVishnya@tempo bro tempo matching businesses with lenders and settling draws in stables under a second is actually useful infra not just another narrative
SHE PULLED A JERSEY OVER HER HEAD AND THE NUMBER CAME OUT STRAIGHT
900k views. 50k likes. 1420 comments. 12 seconds.
filmed next to a black suv on a bright day, "best jersey" on screen. she pulls a teal miami jersey on over a white top, the cap comes off for it and goes back on, she fixes her hair and slings a bag over her shoulder
why it works:
→ fabric going over a head is one of the hardest things to fake. the jersey hides the face, then has to hand it back unchanged, and here it does
→ the order is right: arms through, hem pulled down, and only then do the lettering and the 3 settle flat. print reads once the fabric is taut, which is how a real jersey behaves
→ the camera stands at the car door, handheld, with the door edge cutting into the frame. that is where a friend holding the phone would be
→ hard midday sun splits everything into highlight and shade, and that contrast swallows small texture drift
→ one task from start to finish: getting dressed. no plot, no second location
the takeaway: pick a motion people have watched a thousand times. they know exactly how it should go, so when it goes that way, they stop checking
single-event shots like this are what seedance 2.5 does best, and you can run it inside Picsart