We built a synthetic electorate — one million personas, one sovereign Korean LLM, a two-dimensional ideology space — and locked its forecast for the June 2026 nationwide local elections before the vote. It called 15 of 16 governor races and 192 of 223 mayoral races, with zero opinion polls anywhere in the pipeline. The method, written out with the actual math — plus the one structural flaw that almost sank it, and the fix.
Jul 2026·AI
Demos: the million-persona model that called Korea's local elections
We built a synthetic electorate — one million personas, one sovereign Korean LLM, a two-dimensional ideology space — and locked its forecast for the June 2026 nationwide local elections before the vote. It called 15 of 16 governor races and 192 of 223 mayoral races, with zero opinion polls anywhere in the pipeline. The method, written out with the actual math — plus the one structural flaw that almost sank it, and the fix.
SIDH was one of the last key-exchange schemes standing in NIST's post-quantum competition. Then a classical laptop recovered its secret key in an hour. How it works, the one formula that forces it to leak, and the two-dimensional trick that breaks it — with the actual math — plus a challenge so you can watch it fall.
Jun 2026·Crypto
Sike: the post-quantum scheme that fell to a laptop
SIDH was one of the last key-exchange schemes standing in NIST's post-quantum competition. Then a classical laptop recovered its secret key in an hour. How it works, the one formula that forces it to leak, and the two-dimensional trick that breaks it — with the actual math — plus a challenge so you can watch it fall.
Three years after building an image-copyright framework as a high-schooler, a look at what held up, what didn't, and where the current wave of adversarial-perturbation research is actually pointed.
Apr 2026·Essay
Poisoning the well: why image-generators keep eating copyrighted data.
Three years after building an image-copyright framework as a high-schooler, a look at what held up, what didn't, and where the current wave of adversarial-perturbation research is actually pointed.