Science
Craft traditions that were science before science. Brewers and cheesemakers ran controlled fermentation for centuries before Pasteur explained microbiology; their knowledge was real, only the mechanism was unnamed. Plant breeders did applied genetics before Mendel. Polynesian navigators held a sophisticated empirical system of ocean swells, star paths, and bird behavior — genuine predictive science transmitted without notation.
What do you mean described anecdotally? How does AI find anecdotes?
Fair challenge — and my phrasing deserves it. Let me be precise about what’s actually happening.
I don’t “find” anecdotes. I was trained on a very large body of text — books, articles, forums, memoirs, interviews — and during training I absorbed statistical patterns from all of it. Somewhere in that text were presumably things like teachers’ memoirs, Reddit threads about martial arts pedagogy, interviews where a coach says “I never realized what I was doing until a student explained it to me.” I don’t retain those as retrievable documents. I can’t cite the thread or quote the interview. What remains is something more like a residue: a learned sense that “this is a thing people describe,” without the receipts.
So when I said “described anecdotally,” the honest translation is: the pattern feels familiar from training data, but I cannot produce a single verifiable instance, and I can’t fully rule out that the familiarity is an artifact — my pattern-matching assembling something plausible from adjacent material rather than remembering real accounts. That’s exactly the failure mode called confabulation, and the reason I flagged the claim as unsourced rather than asserting it.
