Robin Salen

ZK Engineering at @0xMiden. Previously @Toposware, @0xPolygon. Building Plonky3 recursion. Maintainer of Plonky3. Opinions are my own.

[Robin Salen @RobinSalen]

🚀 New Plonky3 release just dropped.

This is probably our most impactful and ambitious release so far:

- MUCH faster lookups
- High-arity folding
- N-ary Merkle trees + Merkle caps
- Major Poseidon2 optimizations
- Poseidon1 support
- And many more…

Let’s break it down 👇

[Francois Garillot @huitseeker]

Google’s latest Shor resource estimate: secp256k1 ECDLP (and ECDSA over that curve) may be breakable with <500K physical qubits under specific assumptions. They proved the resource estimate for a key subroutine with a ZKP rather than publishing the attack circuits. (Yes, really - a ZKP to support the claim without disclosing the attack. Interesting times.)

I’ve written about Q-day timing before. The idea was that law & business cycles alone justified working on quantum resistance without trying to predict Q-day.

But you also don’t need to predict the date if the cryptographic primitive your proof system depends on isn’t the one Shor is breaking. Miden’s proof system is STARK-based - hash functions, not elliptic curves.

[Remco @recmo]

Using Groth16 to prove the existence of a quantum advantage in elliptic curves is the only post-quantum secure use of Groth16 :)

[Robin Salen @RobinSalen]

Such a great work! 🔥

I’m excited to see what else we can pull from autoresearch to push ZK forward. Super promising direction!

[Barna @realbarnakiss]

I implemented zk-autoresearch, based on Karpathy's autoresearch loop, on a production ZK prover, Plonky3.

Soundness review by a Plonky3 engineer is pending before I treat these as final. The methodology finding is already clear, preliminary results below.

Target: Plonky3's NTT implementation — the inner loop of proof generation, already heavily optimized by expert ZK engineers. If the approach doesn't work here, it doesn't work anywhere.

Hardware: Hetzner CCX33, AMD EPYC, AVX512, 8 cores.

Model: I used Claude Sonnet 4.6 deliberately, Opus would have marginal gains at significantly higher cost per iteration. For a loop running potentially 100s of times in future experiments, that tradeoff matters.

74 iterations. Fully autonomous by design, but in this first experiment 2 adjustments were made to the setup (at iterations 5 & 10) to nudge the agents to be more decisive.

- Raised MAX_TOKENS from 8192 to 20000, and added "you must always make a change" as the agent kept hitting the token limit. This unlocked improvements at iterations 6 and 9.
- Added near-miss display in the history prompt, showing reverted experiments within 1.5% as combination candidates. This set up iteration 21, where the agent revisited a failed idea that now worked because the surrounding code changed.

Iteration constraints:
- Each iteration ran correctness tests to prevent faulty proofs. Note: during the run these were compile-level checks; post-run correctness was confirmed via full end-to-end ZK proof generation and verification with Radix2DitParallel on BabyBear (10 tests, all passing).
- Agents were structurally prevented from touching FRI or other soundness-critical components — only dft/src/ and baby-bear/src/ were writable.

3% faster at the target size (2^20) during the experiment. Post-experiment benchmarks across 2^14 to 2^22 showed the optimizations generalized better than expected, particularly at the extremes (see image). The agent only optimized for 2^20.

The known issues (short history window causing agent amnesia, wasted tokens on repo exploration, correctness test targeting wrong package) meant the last improvement was found at iteration 21. Round 2 with these fixed should yield a more consistent staircase pattern over 100 iterations.

All gains came from the agent finding redundant work in the hot butterfly loop: precomputing products, hoisting broadcasts, skipping multiplications by 1. Pure implementation-level work, no algorithmic changes.

6 improvements in 74 iterations. 57 regressions. The full experiment log with every diff, benchmark result, and agent reasoning is auditable.

The agent that found these improvements is not a ZK expert. It reasoned about Rust and Montgomery arithmetic from first principles and found real optimizations in code already written by expert engineers.

ZK has been underexplored for agentic optimization because people worry about agents breaking proof soundness. The concern is real but misapplied here, all 6 changes are mathematically equivalent transformations, verified by end-to-end proof generation and verification. (Soundness review by a Plonky3 engineer is pending)

Round 2 is being prepared with the known issues from Round 1 fixed. Full findings and code will be open sourced after it completes.

If you are ZK team and want to run this, feel free to DM me.

Inspired by @karpathy autoresearch pattern. First known application to a production ZK prover.

[ulrich.haboeck @UHaboeck]

Quick announcement: After long and heavy suffering :) the S-two white paper is finally out:
https://t.co/8WyVwoPE6h

Although nothing new in regard to the basic principles (a circle STARK, etc.) the white paper yet contains several details of broader interest:
- A formal description of the flat AIR circuit model (used by several contemporary zkVMs)
- A thorough soundness analysis of multi-table proofs: If one does not use "lifted" FRI, taming the soundness error turns out to be more sophisticated as expected. We introduce the notion of "cross-domain correlated agreement", and show that multi-table FRI satisfies this property.
- A discussion of adjusted conjectures, which takes into account the recent boost of proximity gaps counter examples. We believe that it is plausible to hope for acceptable list- and line-decodability properties up to the information-theoretic barrier, the Elias bound.

Thanks for all the help from the StarkWare team, and in particular to Dmitry Krachun for the many helpful discussions around his counter example.

[Robin Salen @RobinSalen]

I am super thrilled to share that I am joining the @0xMiden team today!

I've been following their work for a while, and it's an honor to be part of their amazing engineering workforce.

To the moon! ✨

[Thomas Coratger @tcoratger]

We are starting to upstream WHIR implementation to Plonky3!

Multilinear primitives have been merged!

Feel free to reach out / open issues / PRs if you want to help or to improve the implementation!

https://t.co/RF9sWWs5LK

[Robin Salen @RobinSalen]

@fede_intern Thanks Fede!

[Robin Salen @RobinSalen]

🚀 New Plonky3 release just dropped.

This is probably our most impactful and ambitious release so far:

- MUCH faster lookups
- High-arity folding
- N-ary Merkle trees + Merkle caps
- Major Poseidon2 optimizations
- Poseidon1 support
- And many more…

Let’s break it down 👇

[Robin Salen @RobinSalen]

Come check the release and build on Plonky3! 👇
https://t.co/5XYQtVjRfM

For builders needing support in migrating their codebase to v0.5.0, we have a Plonky3 telegram channel: https://t.co/IUADYz5Yes

[Robin Salen @RobinSalen]

This release also includes many other improvements across the AIR API, FRI, matrix ops, and prover internals.

Huge thanks to everyone contributing to Plonky3 ❤️

[Robin Salen @RobinSalen]

@tcoratger So we’re not even waiting for new releases now? 🙃

[Etienne Campion @EtienneCampion]

« Il procédait à la discrimination des élèves musulmans. Je vais le dire haut et fort. »

Ces mots, d’une gravité extrême, à l’ouverture du procès en appel des complices de l’assassin de Samuel Paty, sont ceux de Francis Vuillemin, ancien avocat de Maurice Papon et aujourd’hui conseil d’Abdelhakim Sefrioui, prédicateur islamiste dont le militantisme a contribué à la mise à mort de Samuel Paty.

Accuser Samuel Paty, a posteriori, de discrimination religieuse, c’est non seulement proférer une contre-vérité grave, mais prolonger l’entreprise de diffamation qui l’a désigné à la vindicte et conduit à son assassinat. C’est salir la mémoire d’un homme mort pour avoir exercé son métier, au nom d’une stratégie de défense cynique où l’on sacrifie l’honneur d’un mort pour exister médiatiquement et flatter les réseaux islamistes les plus radicaux.

La plaidoirie s’annonce ignoble.

[Jim Sciutto @jimsciutto]

Statement from the family of Alex Pretti, Michael and Susan Pretti: 

“We are heartbroken but also very angry. 

Alex was a kindhearted soul who cared deeply for his family and friends and also the American veterans whom he cared for as an ICU nurse at the Minneapolis VA hospital. Alex wanted to make a difference in this world. Unfortunately he will not be with us to see his impact. I do not throw around the hero term lightly. However his last thought and act was to protect a woman. 

The sickening lies told about our son by the administration are reprehensible and disgusting. Alex is clearly not holding a gun when attacked by Trump’s murdering and cowardly ICE thugs. He has his phone in his right hand and his empty left hand is raised above his head while trying to protect the woman ICE just pushed down all while being pepper sprayed. 

Please get the truth out about our son. He was a good man. Thank you”