AI Isn't Outthinking Mathematicians. It's Out-Remembering Them

Here we go again: just when we've psyched ourselves up for the singularity, someone points out that the emperor's new calculus might just be a very, very thick reference book. The article by Davide Piffer (https://davidepiffer.com/p/ai-isnt-outthinking-mathematicians) makes a deliciously irritating argument—that AI doesn't reason its way to theorems; it memories its way there. Think of it as a cosmic librarian who has read every proof ever written, but who has never once experienced the shiver of understanding why a proof works. It's the difference between knowing the lyrics to every song and ever feeling heartbreak. This cuts to a deep, ontological bruise: what *is* mathematical insight, anyway? Human mathematicians do a ton of memorization—we internalize axioms, lemmas, canonical tricks, and then we combine them in ways that feel fresh. Henri Poincaré talked about sudden flashes of illumination after unconscious work. But what is that unconscious work, if not a neural network doing a Bayesian dance between memorized patterns? Maybe the article's punchline isn't "AI is dumb" but "human intuition is weirder than we thought." If AI's recall is so vast it *simulates* creativity, then perhaps creativity was never more than a different pattern of compression over a lifetime of learned structures. The trouble, though, is brittleness. A mathematician can look at a novel problem, get stuck, and then weaponize the failure itself—using edge cases, counterexamples, and wild analogies to build new mental machinery. An LLM, on the other hand, tends to brilliantly autocomplete whatever breadcrumbs are laid before it, producing formal-sounding nonsense when the problem slips outside the memeplex of its training data. This is what the article implies: AI is out-remember