Posts
2026
What Would It Mean If You Could Cheat? Everything else
On computation, quantum mechanics, the structure of difficulty, and what a universe that allowed shortcuts would have to look like
The Michael Scott Paper Company of AI AI
I just have to wait out you
Bitter-Lessoning neat latent structures through self-study AI
Claude's Constitution
San Francisco Everything else
Ithaka, California
Rationalism Everything else
It will drive you mad (and it's boring)
Books I read (am reading) in 2026 Everything else
I'm going to be better this year now I'm not a founder
2025
Anna Karenina and the arbitrary party Everything else
From San Francisco
Your MLEs are brilliant, but you're giving them the wrong job AI
A letter to the C-suite
The bitter lesson of LLM evals AI
Scaling care beats scaling compute
Five hinge questions that decide whether AGI is five years away or twenty AI
For people who care about falsifiable stakes rather than vibes
Born on third base Everything else
The Case for Inheriting Nothing and Building Everything
One-dimensional vs multi-dimensional features in interpretability AI
How to stop conflating with 768 dimensions
LLMs are really good at k-order thinking (where k is even) AI
You still need to tell a language model you want to cure cancer before it can help you cure cancer.
Information bounds in quantum gravity Everything else
How information theory links quantum mechanics and general relativity
Books I read (am reading) in 2025 Everything else
A dynamic list
2024
The nihilism of NeurIPS AI
Some thoughts on our future in AI research
Can quantised autoencoders find and interpret circuits in language models? AI
Using VQ-VAEs and categorical decision trees to do automatic circuit identification in LLMs.
Learning compressed representations and GPT-5 speculation AI
Why language models probably get too much from the abstraction we give them for free
2023
How I'd build the future of news AI
Notes from Rake
What is the future of Rake? AI
Something I sent to Kirks when we were building