Hyperstax⚡

Friendly reminder: Blockchains tried to reinvent finance. Quilibrium is reinventing the internet. $quil @QuilibriumInc

A lot of the "private" AI options out there are lying to you. Look under the hood. Use the dev tools on the browser to see what is actually sent. It's plaintext. What they're actually doing is _promising_ they won't look. We don't do that. See for yourself. We have an inspector tab so you can see the actual traffic data, but you can confirm it in the browser. klearu-demo.qstorage.quilibrium.com Don't trust. Verify.

MetaVM has been released. Prove execution of RISC-V, EVM, and Solana sBPF in ZK. Supports Quilibrium's BLS48-581 and Ethereum's BLS12-381 natively, without needing a GPU. MetaVM's RISC-V compatibility is the first full ZK RV64IMAC instruction set – you can run Linux in MetaVM, and prove everything that happened within the VM session, prove a block on Ethereum, or a slot on Solana, and emit proofs compatible with either Q or Ethereum. github.com/QuilibriumNetw…

Today, we are publishing one of the side tracks of research ongoing with Q, our E2EE ML training and inference library, klearu: github.com/QuilibriumNetw… SLIDE proved that hash tables can beat GPUs at training deep networks. Further works compounded on this, and Klearu is the first native Rust implementation built on top of this research, extending it to LLM inference, sparsity prediction, and private two-party computation. In the current days we're seeing deeper trust being placed on AI, while the largest of providers are collecting this data for the purpose of not only training, but also advertising, or even selling this data to others. The risks grow worse with every passing day. The majority of AI research for private AI exists in the form of using TEEs – but we've seen time and time again that using TEEs for privacy is disastrous, guaranteed to leak, and even by it's name, is a massive requirement of trust. Outside of this, other private AI looks towards FHE. We know, at least for the near future, that FHE cannot perform at a speed high enough to be generally useful. So instead, we adopted 2PC, with flexible security configurations, where users can be assured that their requests remain private. The majority of these research projects have strictly an output of papers, with no or limited real world instances of their use. Klearu's implementation is available now, with simple instructions for developers to try it out.

Today, we are publishing one of the side tracks of research ongoing with Q, our E2EE ML training and inference library, klearu: github.com/QuilibriumNetw… SLIDE proved that hash tables can beat GPUs at training deep networks. Further works compounded on this, and Klearu is the first native Rust implementation built on top of this research, extending it to LLM inference, sparsity prediction, and private two-party computation. In the current days we're seeing deeper trust being placed on AI, while the largest of providers are collecting this data for the purpose of not only training, but also advertising, or even selling this data to others. The risks grow worse with every passing day. The majority of AI research for private AI exists in the form of using TEEs – but we've seen time and time again that using TEEs for privacy is disastrous, guaranteed to leak, and even by it's name, is a massive requirement of trust. Outside of this, other private AI looks towards FHE. We know, at least for the near future, that FHE cannot perform at a speed high enough to be generally useful. So instead, we adopted 2PC, with flexible security configurations, where users can be assured that their requests remain private. The majority of these research projects have strictly an output of papers, with no or limited real world instances of their use. Klearu's implementation is available now, with simple instructions for developers to try it out.

x.com/i/article/2024…