About

An open notebook for AI research.

lettuceresearch is where I publish my work the moment it's ready — half-finished ideas, negative results, and the occasional paper. It's a single-author archive: I write it, anyone can read it, and the discussion is open to everyone.

lettuceresearch

Independent AI research archive

lettuceresearch is a single-author archive of AI research and notes — work on the inside of language models: how they represent information, where their behavior comes from, and how to keep capable systems aligned with the goals we actually have.

Publishing in the open keeps the work honest and invites sharper feedback than any review cycle. If something here is wrong, the discussion threads are the fastest way to say so.

Focus areas

PyTorch for LLMs

A hands-on, interactive series building up the machinery of modern language models.

Foundations

Hands-on explainers of the machinery underneath modern ML.

Interpretability

Reverse-engineering the circuits inside language models.

Alignment

Keeping capable systems robust to the incentives we give them.

Agents

Self-improving loops, tool use, and emergent behavior.

Retrieval

Giving models longer, sharper, more reliable memory.

Efficiency

Recovering full-model quality at a fraction of the compute.

Multimodal

Grounding language in perception and physical space.

How this archive works

01
Open to read
Every paper, note, and figure is free to read — no account, no paywall.
02
Open to discuss
Anyone can join the discussion under any paper. Corrections and pushback are welcome.
03
Uploads are owner-only restricted
Research is published by one author — this is a personal archive, not a community repository. There is no public submission.