Andrej

Nanochat. I built a way to train your own ChatGPT for fifty dollars.

Autoresearch. I built an AI that experiments on machine learning models while you sleep.

Council. I built a tool that asks multiple AI models to answer your questions and debate the best response.

NanoGPT. I made the simplest code to train your own AI language model.

LLM101n. I built a course that teaches you to create an AI storyteller from scratch.

LLM.C. I built AI model training that runs in plain C without bloated dependencies.

Reader3. I made a simple ebook reader that works with AI models to read books together.

Job Market Visualizer. I built an interactive map of every US job with government data.

Rendergit. I built a tool that flattens GitHub repos into searchable single pages.

Karpathy's Blog. I write a blog explaining AI and machine learning concepts clearly.

Neural Networks Zero to Hero. I made a course teaching neural networks from the ground up.

HN Time Capsule. I built a tool that grades decade-old predictions with hindsight.

Rustbpe. I made a fast, simple tool for training AI tokenizers.

Llama2.c. I made an AI model you can train and run in one C file.

MinBPE. I wrote clean code for the text-splitting algorithm all AI models use.

Build NanoGPT. I made a video course teaching how to build GPT-2 from scratch.

MinGPT. I built GPT from scratch in 300 lines of readable code.

NanoGPT Lecture. I published the code from my neural networks video course so people can learn by building.

Makemore. I made a tool that generates realistic new text by learning from examples.

Calorie. I made a simple calorie deficit tracker for weight loss.

Random Fun. I collected my experimental notebooks and scripts in one place.

Karpathy. I shared my thoughts and experiments on how AI models work.

Arxiv Sanity. I built a website that tames the overwhelming flood of research papers.

Arxiv Sanity. I made a tool that learns which research papers interest you and recommends similar ones.

Cryptos. I built Bitcoin from scratch in Python to show how it works.

Micrograd. I built a tiny AI learning engine you can read in an afternoon.

LeCun 1989 Repro. I reproduced Yann LeCun's 1989 handwritten digit recognition experiment.

Deep Vector Quantization. I wrote training code for turning images into learnable symbol sequences.

SVM.js. I built a machine learning sorter that runs in the browser.

NeuralTalk2. I built software that automatically writes captions for images.

tsneJS. I wrote a tool that visualizes large datasets as interactive maps.

Paper Preview. I made scripts that turn research papers into browsable web pages.

EigenLibSVM. I built a C++ tool for training machine learning models on matrix data.

Ulogme. I built software that tracks your computer activity and shows you a timeline of your day.

Agent Trainer. I wrote reinforcement learning agents for simulated environments.

Forest. I built a random forest tool for MATLAB with pluggable learners.

NeuralTalk. I built an AI that describes images with sentences.

CharRNN. I built an AI that learns to write by studying text samples.

Gitstats. I built a tool that shows your team's recent work across all code branches.

ScholarOctopus. I made a map of 7000 academic papers from major conferences.

REINFORCEjs. I built a learning system where programs improve through trial and error.

Density Estimator. I implemented a density estimation model in about 150 lines of code.

RecurrentJS. I built AI models with memory in the browser.

ResearchLei. I made a filing system for research papers you can actually search.

Covid Sanity. I built a searchable library of COVID-19 research papers organized by similarity.

ForestJS. I made a machine learning tool that learns patterns and makes predictions in JavaScript.

NotPygameJS. I made a toolbox for building games in web browsers.

Research Pooler. I built a tool that automatically finds related research papers for you.

Find Birds. I built a tool that finds Twitter accounts you should follow.

Normalizing Flows. I built an educational tool for understanding how normalizing flows transform data.