I saw the message from test_user in market chatter (I don’t usually use this function) about the article in FT.com.
Anyway, I check out this article which compares quants and AI Lab and find this particularly interesting
But in both worlds the real edge lives in things that never hit arXiv: which sliver of data you actually trust; the odd combinations of regularisation, learning rates, and curriculum schedules that converge; the deployment hacks that make the system stable at scale. Quant finance has always been explicit about this. The famous unwritten rule at Renaissance Technologies was that they should never publish anything even vaguely useful — including failed experiments. When a couple of RenTech researchers left for Millennium in 2003, a New York court enforced their non-competes in nuclear fashion. New York law, plus the pod-shop ecosystem, treats models and data pipelines as core trade secrets.
Regards
James
EDIT: I am glad there is still a lot of sharing of experience in P123 and thanks to test_user again for the heads up to this interesting FT article.