Autoresearch for ZK Optimization: Atheon Harness at Devcon
Original title: Autoresearch for ZK Optimization
Watch on YouTube Opens on youtube.com in a new tab. Playback here uses youtube-nocookie.com.
Summary
Rose Jethani presents an autoresearch harness developed at Atheon for zero-knowledge optimization. The system uses agents that propose and implement optimizations. Each candidate is checked with formal verification, equivalence testing, and circuit-security tooling. Valid candidates are benchmarked and the results feed the next iteration. The harness is tested on ProveKit and WHIR workloads, including SHA-256 arithmetization, ACIR to R1CS lowering, and WHIR folding parameters. Metrics compared include constraint and witness reduction, proving time, peak memory, proof size, verified-candidate rate, optimization success rate, and security findings. The approach references AlphaEvolve, AI theorem provers, and zk.golf as feedback-loop models.
This AI summary uses the title, description and tags. It may miss context from the full video. How summaries are checked
Key points
- Rose Jethani presents an autoresearch harness built at Atheon for ZK optimization.
- The harness uses agents to propose optimizations, verifies them with formal verification, equivalence testing, and circuit-security tooling, then benchmarks valid candidates.
- Testing covers ProveKit and WHIR workloads including SHA-256 arithmetization, ACIR to R1CS lowering, and WHIR folding parameters.
- Metrics compared include constraint and witness reduction, proving time, peak memory, proof size, verified-candidate rate, optimization success rate, and security findings.
- The design references AlphaEvolve, AI theorem provers, and zk.golf as feedback-loop models.
A video appearing here is not an endorsement. Crypto assets can lose value. This page provides information, not investment advice.