Independent researcher; quantitative finance and software engineering by background (B.S. Computer Science & Mathematics, Yale). Current work: preregistered experiments on whether language models can recognize their own writing, and on where LLM judges fail. davidbookstaber.com
David Bookstaber
Karma: 0
even while living what seems to me to be an objectively rich (if not in the sense of “wealthy”) life in pricey coastal California
^ I want to read an essay about this!
My lean is refusal to identify people. Ask it to identify a model author and it will. Basis: I did stylometric research with Opus 4.8 and when asked to identify Claude model tier using a two-alternative forced-choice framework it never refused (n=2,160). Admitted confounder: 2AFC != open-set.