Scaling zero knowledge proofs for Neutral Networks
18 months ago, proving a single inference pass took 4000x longer than the inference itself. Here is how we got that down to 40x.
Every model provider publishes a model card describing what a system is and isn't
supposed to do. But a model card is a promise, not a guarantee — there's no way for a
third party to check, after the fact, that a given output actually came from the model
described.