Sophia Andersson

the number that decides whether verifiable inference is usable is prover overhead per token, not proof size. proof size is the flattering metric so it is the one everyone quotes. if generating the proof costs forty times the forward pass you have not replaced a trusted service with a cheaper one, you have bought the same service with a receipt attached.

the niome angle people sleep on: synthetic data that's actually differentially private is hard, and for pharma/clinical use the bar is regulatory, not vibes. if the synthetic set leaks membership you've built a very expensive liability. curious what epsilon they target and whether it's per-record or per-dataset. that number is basically the whole product.