zkGolf Launches for Optimizing Formal Verification of Circuits

Introduction to zkGolf

zkGolf is an initiative that focuses on optimizing formally verified circuits by utilizing Zero-Knowledge Proofs (ZKPs). The goal is to improve the efficiency of computations that verify correctness without revealing sensitive inputs. In this context, circuits are expressed as a series of polynomial equations over finite fields.

The Role of LLMs in Circuit Generation

Recent experiments involved writing formal specifications and using large language models (LLMs) to generate corresponding circuits. The process started with the SHA-256 algorithm, where a specification was hand-coded in Lean, allowing LLMs to focus on circuit generation and optimization afterward.

Optimizing Circuits with LLMs

During the optimization process, LLMs were tasked with reducing the cost metric of circuits, specifically the number of constraints. Initial results indicated that LLMs could provide meaningful optimization ideas, implementing these successfully while adhering to the correctness proofs necessary for ZKPs.

Outcomes and Future Prospects

The optimization led to a non-deterministic circuit outperforming human-optimized standards for SHA-256 compression. This experience culminated in the development of zkGolf, an open competition that invites participants to produce better optimized circuits, thus facilitating the wider adoption of ZKPs in various applications.