Wyatt Benno

@wyattbenno

9 Following

417 Followers

Wyatt Benno pfp

I am a big fan of client side proving and was checking out STWO. I think the marketing people don't get it AT ALL and often mislead 😱 Marketing: "not only scalability but also privacy—increasing economic freedom by enabling private transactions on Bitcoin and Ethereum at massive scale" https://starkware.co/blog/s-two-prover/ Dev docs are more honest: https://zksecurity.github.io/stwo-book/why-stwo.html#:~:text=As%20of%20the%20time%20of,to%20infer%20the%20witness%20values "As of the time of this writing, Stwo does not provide the "zero-knowledge" feature. "Zero-knowledge" here refers to the fact that the proof should not reveal any additional information other than the validity of the statement, which is not true for Stwo as it reveals to the verifier commitments to its witness values without hiding them by e.g. adding randomness. This reveals some information about the witness values, which may be used in conjunction with other information to infer the witness values."

0 reply

0 recast

5 reactions

Wyatt Benno pfp

I have heard a few people say : "there will only be one or two companies that win the ETH realtime proving race". But @vitalik.eth points out some interesting facts. I would add this to 3: Most provers use special hardware and are extremely memory hungry. Folding scheme are the most direct way to decentralize and democratize ZKP proving — it is the best technique that we have to allow provers of any size to join and prove. In the end, I think there will be ZERO companies left and just networks of people. Most likely they will be using memory efficient folding scheme in a collaborative setting to achieve realtime proving at scale. https://x.com/VitalikButerin/status/1925050155922862526

0 reply

0 recast

5 reactions

Wyatt Benno pfp

We are proud to open-source a high-performance MSM for BN254 curve in WebGPU. It runs on Windows, macOS, AMD chips, many mobile devices, and in the browser. Optimized for speed, portability, and massive parallelism. Check it out: https://github.com/ICME-Lab/msm-webgpu __ The implementation follows the cuZK paper closely and is highly optimized for GPU workflows. WebGPU is not well documented and hard to test - thank you @zk_albi for the painstaking hours spent making this come to life :) https://eprint.iacr.org/2022/1321.pdf Want to know what we are using it for? 👇 1/2

1 reply

0 recast

24 reactions

Wyatt Benno pfp

Zero-Knowledge Machine Learning is Sus, But It Won’t Be for Long. In this post I explore the costs of ML for ZKP proving — how GKR, sumcheck protocol and 'sparsity' can help! Particularly in the the elliptic curve or lattice setting. Moreover some insight from our work with lookup arguments and JOLT for non-linearities. https://www.novanet.xyz/blog/verifiable-ai-with-zero-knowledge-proofs

0 reply

0 recast

6 reactions

Wyatt Benno pfp

We have been doing some work on SoTa zkML that has me much more bullish on the use-case. Should be a fun talking point at zkSummit 13 😅

0 reply

0 recast

3 reactions

Wyatt Benno pfp

Homomorphism is a mathematical super power that EC and lattice play with. Great listen on lattice 🐦‍🔥 https://x.com/zeroknowledgefm/status/1917573411502461062?s=46

1 reply

0 recast

7 reactions

Wyatt Benno pfp

ZK land is a bit like the current wars: direct current (DC) vs alternating current (AC). FRI is our DC - backed by early supporters and initially dominant. Folding schemes are our AC - elegant and efficient at scale. DC was Edison's baby and had early success, but Tesla's AC ultimately transformed power distribution. If history rhymes in cryptography Michael Arrington is cast as Westinghouse 😂🏴‍☠️. https://x.com/arrington_cap/status/1917569694011207867?s=46

1 reply

0 recast

5 reactions

Wyatt Benno pfp

Cool new paper on fast accumulation with hashes. First time thinking about moving away from homomorphic EC/Lattice for small space provers. https://eprint.iacr.org/2025/753

0 reply

0 recast

6 reactions

Wyatt Benno pfp

SNAIL: GKR that replaces sumcheck with batched zerocheck. https://eprint.iacr.org/2025/728.pdf

1 reply

0 recast

7 reactions

Wyatt Benno pfp

Getting ready to present in NYC on Sat.. "Verifiable Compute Is Eating The World." At: https://lu.ma/kaqoydqn

0 reply

0 recast

4 reactions

Wyatt Benno pfp

A zkVM that only runs on massive, custom rigs? Centralizing. JOLT + folding gives you a fast zkVM that runs anywhere— from beefy servers to edge devices. That’s decentralization. +Privacy w/ HyperNova folding. Necessary and sufficient.

0 reply

0 recast

1 reaction

Wyatt Benno pfp

Introducing 'Proof Parties' - a platform for games and competitions demonstrating the state of the art in ZKP. At ETH Denver I heard a bunch of : "Why is ZK important?" "What can I do with it?" "Show me!" 1/4 thread. Try it: https://zkp.fun/ My blog about it: https://icme.ghost.io/proof-parties-zero-knowledge-proofs-with-friends/

4 replies

8 recasts

59 reactions

Wyatt Benno pfp

zkECDSA talk on SoTa. Noir does well! But there are a few tricks that can improve it a lot. https://blog.hyle.eu/benchmarking-in-browser-p256-ecdsa-proving-systems/

0 reply

2 recasts

5 reactions

Wyatt Benno pfp

Just so we all know we can keep privacy and speed with delegating proving w/ holographic snark 😊😎. Logs: `Time elapsed for witness-related (client-side) part of proving w/ delegated Spartan is: 346.279ms Time elapsed for delegated part of proving w/ delegated Spartan is: 6.68s`

0 reply

0 recast

4 reactions

Wyatt Benno pfp

So what's new in folding implementations / client-side ZK? So many cool things. But here is a quick update. 1. We made an on-chain verifier that works in ~750,000 gas.. based on the work from PSE. Folding + cheap on-chain verification allows big / aggregated computations to trigger smart contracts cheaply. Read: https://www.novanet.xyz/blog/any-computation-is-verifiable-with-cheap-on-chain-folding-verifiers 2. 3x speed up seen from preliminary NeutronNova work.. ported to zkEngine this means 3 minutes proofs in browser are 1 minute proofs. All pre-webGPU. 🤯 Recent work also moves public param gen time from 24.3s to 3.2s. We will be releasing a concept called "Proof Parties - have fun proving with you friends" in a week or two so.. Stay tuned ⏱️ PR: https://github.com/microsoft/Nova/pull/368

0 reply

18 recasts

107 reactions

Wyatt Benno pfp

SOTA - Lattice based folding: https://x.com/srinathtv/status/1892701954939490697

1 reply

0 recast

6 reactions

Wyatt Benno pfp

Cool progress on Lattice based folding🔥 https://eprint.iacr.org/2025/247.pdf

0 reply

0 recast

5 reactions

Wyatt Benno pfp

Technical founders — embrace your inner polymath.

0 reply

0 recast

4 reactions

Wyatt Benno pfp

If A 'ZK'-Prover Network Asks For Your Data. Don’t Give It to Them 🧐 In this post I discuss delegated proving, CoSnarks, zkMPC, zkFHE, schemes that support fast client-side proving, while offloading the expensive prover work to untrusted provers. https://blog.icme.io/zk-prover-networks-want-your-data-dont-give-it-to-them-2/ I argue that prover networks and zkVM should support privacy as a default and the general structure of prover networks.

1 reply

1 recast

2 reactions

Wyatt Benno pfp

China’s eery silence? Not responding with retaliatory tariffs seems to be a good response. 10% brunt paid by US consumers while domestic consumers in China are completely unaffected. Moreover if the yuan becomes weaker as a result of a trade war their products seem more desirable, increasing the lot of many in China. US seems like unreliable trading partner.. etc etc. US side needs a clear strategy..

0 reply

0 recast

2 reactions