According to Mars Finance, Vitalik Buterin recently wrote an article explaining how the GKR (Goldreich–Kahan–Rothblum) protocol is used to accelerate ZK proofs, adapting to the “batch × multi-layer” computation structure, significantly reducing the intermediate layer commitments, committing only to the inputs and outputs. The article uses Poseidon2 hash as an example to detail the recursive proof process centered around sumcheck and provides optimizations (Gruen’s trick, linear batching, partial rounds with only cubic leading elements), which can be combined with BaseFold or FRI in polynomial commitment scenarios. The author claims that the actual overhead is about 100 times lower than the theoretical value of traditional STARK, with single-digit level overhead expected, and warns that the Fiat–Shamir challenge needs to guard against predictability risks within the circuit.