The Jalay Bee

Andrej Karpathy Founding member of OpenAI, former Tesla AI director, now educator and independent voice on model training, agents, and the future of software.
Simon Willison Independent developer, co-creator of Django, prolific blogger and tools-builder operating at the frontier of practical LLM usage.

Karpathy: the eval game has lapped the model game
Simon Willison's quiet tool shipping is doing more for indie AI dev than any frontier release

Karpathy: the eval game has lapped the model game
Simon Willison's quiet tool shipping is doing more for indie AI dev than any frontier release

llm.c

No posts yet about llm.c.

autoresearch

AI agents running research on single-GPU nanochat training automatically. A 630-line self-contained repo for autonomous LLM training iteration.

llm-council

Multiple LLMs work together to answer your hardest questions.

nanoGPT

The simplest, fastest repository for training and finetuning medium-sized GPTs.

Timeline

llm.c

2024

May 2024

GPT-2 (1.5B) trained in pure C

Public training reproduction at 1.5B parameters; the "how does this all actually work" reference implementation.
Apr 2024

Initial commit

Karpathy publishes a from-scratch C/CUDA training loop for GPT-2; ~1000 lines, no PyTorch.

2 total events