From a development perspective, if the program is just a SHA2, consisting solely of logic operations, and is easy to implement, custom ZK is definitely more efficient than zkVM. However, when dealing with zkML for instance, which involves complex computations like floating-point arithmetic, non-linear functions, and elliptic curve pairing, custom ZK might not be as efficient.
zkVM can leverage continuation and proof aggregation to parallelize proof generation, significantly speeding up the process. Thus, efficiency should be defined in terms of both build time and run time; focusing solely on run time is insufficient.
For example, before computers, we used calculators, which were faster for simple math problems. But with general-purpose computers, we can perform much more complex calculations far more quickly. Both are efficient in different scenarios, which is why we use both calculators and computers today.
https://x.com/vanishree_rao/status/1817617950624698782 1 reply
3 recasts
12 reactions
0 reply
0 recast
0 reaction
