OpenAI says about ten thousand AI agents solved Navier-Stokes, and Terence Tao worries that mathematics is turning into a production quota. My own AI-assisted math is small and slow, and I keep a dated record of every step. That record, which anyone can check line by line, is what I think is worth setting beside a 10,000-agent announcement.
Posts
Essays by Tyler Orden
I do narrow, checkable math research with AI helpers, such as extending number sequences in the OEIS, the reference database mathematicians use to look up sequences, and I keep a dated record of every step. These essays are about things that happened in that work and what I took from them. For short how-to lessons, see this site's guide.
An AI's explanation for my counting error was wrong, and how it was tested
In August two of my counting programs agreed with each other and were both wrong. The explanation the AI agent running the project wrote down that evening went into my planning files and suggested a paper idea, and it was impossible: it needed a corner of a honeycomb where four edges meet, and every corner there joins three. How a later test ruled it out, where the real cause had been since July, and the test I now run on every explanation for an error.
I was on the wrong journal's website, and what I rechecked afterward
I was halfway through submitting a paper when a checkbox asked me to agree to a required publication fee. The journal I meant to use charges nothing: I was on the site of a different journal with an almost identical name. Nothing was sent, but I had already split my paper into four files to fit that other journal's rules. What I rechecked, and the warning sign I watch for now.