New Developments on the AI Integrity Track: Attestable Announces Completion of a $20 Million Seed Round and Officially Launches Its Core Product.
This round was co-led by Jamin Ball of Altimeter Capital and Yonatan Mandelbaum of TLV Partners, with participation from institutions including Halcyon Futures, Cerca Partners, and Zero Knowledge Ventures. Even more noteworthy is that Vitalik Buterin reshared a tweet from Attestable CEO Yogi on the X platform, praising it as “an impressive piece of work,” underscoring that industry KOLs recognize this direction.
The project’s biggest highlight is that it tackles the long-standing technical bottleneck of “practical zero-knowledge proofs.” Its approach shifts the trust assumption from data centers to an extremely small mathematical verifier: the data center can prove that the approved model, weights, inputs, and strategies indeed produce the corresponding outputs. The proving process neither leaks model weights and private data nor requires private keys for verification, and it also eliminates the need to rerun the model.
In terms of performance, the team discloses that its alpha version achieves a verification speed of 85 tokens per second on a single NVIDIA H100 for the Meta Muse Glimmer 30B model. The proof is compact and also post-quantum secure—practically speaking, it is already close to the bar for commercialization.
The combination of zero-knowledge proofs and AI inference verification has long been viewed as a key pathway to solving the problem of “black-box model trust.” With Vitalik’s public endorsement this time and measurable performance data delivered as well, the next areas worth watching include: whether it can attract more model vendors and cloud service providers to integrate, and whether proof costs can drop further as hardware evolves.
#AI #零知识证明 #Web3
This round was co-led by Jamin Ball of Altimeter Capital and Yonatan Mandelbaum of TLV Partners, with participation from institutions including Halcyon Futures, Cerca Partners, and Zero Knowledge Ventures. Even more noteworthy is that Vitalik Buterin reshared a tweet from Attestable CEO Yogi on the X platform, praising it as “an impressive piece of work,” underscoring that industry KOLs recognize this direction.
The project’s biggest highlight is that it tackles the long-standing technical bottleneck of “practical zero-knowledge proofs.” Its approach shifts the trust assumption from data centers to an extremely small mathematical verifier: the data center can prove that the approved model, weights, inputs, and strategies indeed produce the corresponding outputs. The proving process neither leaks model weights and private data nor requires private keys for verification, and it also eliminates the need to rerun the model.
In terms of performance, the team discloses that its alpha version achieves a verification speed of 85 tokens per second on a single NVIDIA H100 for the Meta Muse Glimmer 30B model. The proof is compact and also post-quantum secure—practically speaking, it is already close to the bar for commercialization.
The combination of zero-knowledge proofs and AI inference verification has long been viewed as a key pathway to solving the problem of “black-box model trust.” With Vitalik’s public endorsement this time and measurable performance data delivered as well, the next areas worth watching include: whether it can attract more model vendors and cloud service providers to integrate, and whether proof costs can drop further as hardware evolves.
#AI #零知识证明 #Web3