On July 23rd, Vitalik Buterin published "Exploring circle STARKs", the article said that in the past two years, the most important trend in STARK protocol design has been the shift to operate on a small scale, which has led to a significant improvement in proof speed. Starkware is currently able to prove 620,000 Poseidon2 hashes per second on M3 notebooks, which means that if we are willing to trust Poseidon2 as a hash function, then the hardest part of making efficient ZK-EVM has been effectively solved. Vitalik believes that through the combination of binary field technologies such as Mersenne31, BabyBear, and Binius, we do feel that we are approaching the limits of STARK's "base layer" efficiency. At this point, I expect the frontier of STARK optimization to shift to the most efficient arithmetic operations on primitives such as hash functions and signatures (and optimizing these primitives themselves for this purpose), recursive construction for more parallelization, arithmetic operations on VMs to improve the developer experience, and other higher-level tasks.