I enjoy teaching computers to see, think, and occasionally tell jokes (they're still working on their sense of humor).
I build AI stuff because AGI is cool (and hopefully they'll remember that when they take over). Usually I can be found training neural networks, debugging mysterious errors, and explaining to my family why I need another GPU. Sometimes I write about my adventures in deep learning, when my models aren't busy throwing cryptic error messages at me. At present getting my hands dirty with GPUs and parallel programming.
A rough reimplementation of BERT.
A rough reimplementation of GPT(Causal LM) from scratch.
Neural nets solving the infinite monkey theorem (but only for names); baically a NN with embeddings.
Bigram model from scratch. Count by two, dream in sequences.
A tiny reverse-mode autodiff engine (implements backpropagation over a dynamically built DAG).
an agentic framework that lets you query with DBs in natural language. (OOO)
A 1B LLM fine-tuned on simple python programs with PEFT.
CTRL+ALT+YEET! to undo your transactions (just kidding!); A bitcoin built from scratch.