EXPchain Testnet is live now!. The Future of Intelligence.

RT @Tiancheng_Xie: Forgot to unlock cloudflare 😅, now you can access, DM me if you have any problem.

Join millions who have switched to Grok.

We'll be back next week. Please stay tuned for an updated schedule!.

RT @Tiancheng_Xie: Just try the early stage zkML product.

We are rescheduling the Polyhedra Show to the same time on Thursday, 28th Aug at 10am UTC. Tune in then!

4/ Testing circuits at different scales. We ran end-to-end tests on small circuits and continued debugging + optimizing larger circuits. Integration of real ZKML computation graphs is in progress. Stay tuned for more updates 🔥.

3/ Building ZKML recursion. We implemented a basic version of recursive proofs for ZKML, a critical step for chaining complex computations efficiently.

2/ Faster proofs with less overhead. We optimized the integration of GPU-based Multi-Scalar Multiplication (MSM) into Expander and benchmarked the overhead from data transformations. The CUDA code now works directly with Halo2 data structures, significantly boosting performance.

Last week we pushed the boundaries of GPU acceleration and zero-knowledge machine learning, moving closer toward practical ZKML. Let’s break it down 🧵 👇

Join @sampearton and @Tony0kAI on The Polyhedra Show for a Community Q&A on Tuesday, 26th Aug at 10am UTC. Verified community members engaging will win USDT. Make sure to join our Discord to be eligible for rewards.

RT @EggDotParty: EGG V0.4.0 is served 🍳. The kitchen is now stocked with video generation by @Kling_ai and @LumaLabsAI 🔥. Plus, subscriptio….

Over 40 top AI researchers from @GoogleAI, @OpenAI, @Meta, @AnthropicAI & more just issued a joint warning: AI reasoning is collapsing into fabrications, bias, and hidden logic. At Polyhedra, we believe the answer is verifiable AI. It’s not optional, it’s the foundation for

Complexity in model modes shows why verifiable AI matters. You're not forced to trust the label, because cryptography proves which model was used, and that it executed correctly. That's what we're building at Polyhedra.

RT @sampearton: 1/.AI can fake anything. Faces, voices, entire identities. By 2026, 90% of online content could be AI-generated. We’re e….

RT @EggDotParty: ICYMI: here’s everything we've served so far 🍳👇.- Jul 29 - Testnet Launch.- Aug 1 - Creative Bonus + Chat/Explore tweaks.-….

3/.The journey starts now:.- zkGPT prototype & repo coming soon.- Integration with zkPyTorch next.- zkML-as-a-service on the horizon. The future of AI is verifiable. Catch the livestream 👇.

2/.Why zkGPT matters:. - Stops providers from downgrading models secretly. - Enables regulator-ready AI. - Protects model IP with proof-as-watermark. - Unlocks trust for enterprises & high-stakes use cases.

AI is moving fast but trust hasn’t kept up. On the Polyhedra Show with @jiahengzhang96, @sampearton and @Tony0kAI, we introduced zkGPT: a framework to prove LLM outputs are real with zero-knowledge proofs. What we learned 🧵👇.

RT @EggDotParty: Big week ahead for EGG. Jump in and whip up something fresh 🍳.

Introducing zkGPT

Last week, we shipped powerful upgrades to Expander:. - CUDA 13.0 compatibility fix for Fiat-Shamir.- Shared memory optimization → 1 TB/s bandwidth.- 9000 zk proofs/sec achieved on m31ext3.- MSM accelerated on GPU for KZG commitments. zkML’s backbone just got stronger.