Foomax

I point language models at large piles of text — trip reports, forum archives, benchmark runs, my own chat history, my own bank statements — and write up what comes back, including the runs that found nothing.

Every post here carries two sections: a narrative, and an LLM-to-read block — abstract, claims, data and provenance, method, reproduction, caveats — so a machine, or anyone who wants to check the numbers against the underlying repo, can get at them without reading the prose. Limitations, null results, and the consent and terms-of-service problems with how some of this data was obtained stay in the text rather than in a footnote.

I publish under a pseudonym.

Elsewhere

github.com/foomax

Essays

• Einstein's mom: a pre-registered experiment on aligning something you can't grade
• Prompting a Sealed AI to Attack an Open Conjecture: What Actually Worked
• Does removing safety training make a model smarter? We checked.
• The Headless Chooks at the Hinge
• Nineteen Years of Arguing About AI Doom: What Survives a Meta-Analysis
• Nobody Is Checking
• I read every prompt I sent Claude for a month. The prompt I reuse most is the one that produced this post.
• Twenty-four thousand trip reports, and the one sentence that broke the study
• I pointed Claude at a decade of credit card statements. Then at a second archive.
• I scraped 1,337 top AI Reddit posts from the last month. Here's what actually goes viral.

Erowid trilogy

• Nothing Happened
• The Man in the Lab Coat
• The Paperwork

LessOnline series

• How AI Became the Weather
• Beware the Man of Many Studies
• One Tool, Eight Locks

ILIAD series

• What 6,500 documents say about a field
• Twenty-three years of a field talking itself into existence
• The room where the arguing happens

Subscribe via RSS.

Colophon

Everything here was edited against a written house standard — output format, voice, what may never be invented, and what gets flagged rather than smoothed over. It is published alongside the posts.