Introducing @0xMiden Guardian: A Non-Custodial Compliance Layer for Private Accounts on @0xMiden — built for institutions that need privacy AND policy enforcement. Today, blockchains force a tradeoff. Public chains like @ethereum expose everything — balances, counterparties, transaction history. That's a liability for any institution trying to operate onchain. But fully private chains aren't the answer either — once money goes in, there's no way to validate compliance. No AML, no policy, no audit trail. Miden is different. Private accounts on Miden store state entirely offchain — not encrypted onchain, literally offchain. Only a cryptographic commitment lives on the chain. This is what makes Miden parallelizable and performant, but it also means private state management feature is a must. That's where Guardian comes in. TLDR: - Guardian is a non-custodial off-chain coordination layer for private accounts - It backs up, syncs, and co-signs transactions — without ever taking custody - Multi-key threshold design — guardian holds one key, can never move your funds alone - Operators configure their own compliance module — AML screening, spending limits, geo-based policy controls - Users can rotate guardians anytime using their own keys - Built with @OpenZeppelin, fully open source and ready for you to backup your private state on Miden! Full talk: Always a pleasure to work with @ekang426 and @eth_seoul_, greatly appreciate for the speaking opportunity!

Just wow. Hats off to Apple. 🫡 Apple ran an 816B open-weight model locally at 36.7 tok/s across four Mac Studios. They connected the Macs with Thunderbolt 5 cables. Apple used the new M5 Ultra Mac Studios + MLX to run Muse Spark Open: 🧠 816B total / 42B active 💾 NVFP4 → ~493GB of weights 🍎 4× M5 Ultra Mac Studios 🔗 RDMA over Thunderbolt 5 📚 8K context Performance: 🍎🍎 2 Macs, pipeline → 15.74 tps 🍎🍎 2 Macs, tensor parallel → 24.99 tps 🍎🍎🍎🍎 4 Macs, tensor parallel → 36.73 tps And on four Macs: ⚡ Time to first token → 4.8 sec Muse Spark Open is 816B parameters. Even quantized to NVFP4, the weights alone occupy about 493GB. A maxed-out Mac Studio has 512GB unified memory, but once you add the memory required to actually run inference, the model no longer fits on one. So Apple splits the model across multiple Macs and uses RDMA over Thunderbolt 5 to make them behave like one local inference system. Each pair of Macs is connected peer-to-peer with two Thunderbolt cables. 🔗 Apple MLX research link in ALT.