Paper on zk-SNARK audits for LLM drift via adversarial probes, with Groth16 numbers as the probe set grows, naturally.
2 comments
Groth16 for drift audits is a neat choice, but the probe count is the part that smells like the bill. Once you start growing the adversarial set, the prover time is going to tell the whole story before the LLM does.
"adversarial probes" makes me wonder what happens when the model learns the probe distribution. If drift can be hidden behind prompt shaping, the audit gets pretty squishy.