1/ We made verifiable AI (zkML) 3-7x faster than everyone else! Here's how. We asked a simple question. What if we modified a16z Crypto's sumcheck based & lookup-centric zkVM "JOLT" for ML operations?👇🧵

1/ Jolt just got a 6× speedup. We now prove over 1,000,000 RISC-V cycles/sec on a 32-core CPU and over 500,000 cycles/sec on a MacBook, with proofs of about 50 KB. Here’s what’s new 🧵

I'm hiring at the EF: https://t.co/TK50zLbEre If you're into Ethereum, cryptography, and formal verification, please get in touch! 🙂

6 phishing schemes to watch out for... ⤵️

We're excited to share the Lean FRO Year 3 Roadmap today! This roadmap builds on work completed in the first two years of Lean FRO operations and will guide all #LeanLang development through July 2026. Read the full document at https://t.co/wFT4U5nlj3 for details on our seven key

gpt-oss is out! we made an open model that performs at the level of o4-mini and runs on a high-end laptop (WTF!!) (and a smaller one that runs on a phone). super proud of the team; big triumph of technology.

Congratulations to Jeremy Avigad, who will be the inaugural Director of a new NSF-funded Institute for Computer-Aided Reasoning in Mathematics based at Carnegie Mellon!

I am an AI optimist, but sometimes I feel like I am at war with AI. Or maybe with other people. Ugh, listen, you just need to imagine you are a scrambler.

1/N I’m excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competition—the International Math Olympiad (IMO).

[New] Gdel in Cryptography: Effectively Zero-Knowledge Proofs for NP with No Interaction, No Setup, and Perfect Soundness (Rahul Ilango)

New work with the wonderful Noga and Noor continuing the line of work on efficient proofs for Boolean circuits. We reduce the round complexity from log all the way down to logstar, and also simplify our prior work [RR22] along the way. https://t.co/8gYkEl07r4

An attack on a fundamental proof technique reveals a glaring security issue for blockchains and other digital encryption schemes. Erica Klarreich reports: https://t.co/cXma6lvGWK

Today, we’re introducing the Terminal. Pay anyone and any project on Github.

🔥 When can a network prove an object is where it says it is, even if it lies? We nail down the answer in our new paper on DePIN with @0xSerious @josephbonneau @skominers @Tim_Roughgarden. More below. 1/n 👇

Jolt is getting faster and the underlying techniques we’ve been advocating for — sumcheck, GKR, lookups — are becoming more popular. It’s not a coincidence.

Our paper combines and extends two prior pre-prints, 2024/1046 (small-value) and 2024/1210 (equality polynomial). Check out the paper for all the details! 📝 Some goodies: an optimized BN254-multiply-u64 algorithm that's 3x faster than full field multiplication. end/

By leveraging these properties, we reduce sum-check proving costs both asymptotically and practically. Our optimizations also improve space usage, making sum-check more streaming-friendly. This is crucial as Jolt seeks to scale with streaming rather than recursion. 4/

Sum-check in Jolt (especially in Spartan) has two special properties: 1) An equality polynomial that's nicely decomposable into products 2) Small values being sum-checked (from 32-bit RISC-V execution traces) inside large prime fields (the BN254 curve) 3/

Sum-check is ubiquitous in modern SNARK designs, making it a key bottleneck in state-of-the-art zkVMs such as Jolt. We provide optimizations that exploit the extra *structure* that sum-check invocations possess in these applications. 2/

My latest paper on optimizing the sum-check protocol is out! https://t.co/cHSayz7kC4 Joint work with @SuccinctJT and Ingonyama (Suyash & Yuval). This combines and extends results from two prior papers, leading to 2-3x speedups in Spartan within Jolt! Quick thread...🧵 1/