One of the fascinating things about AI and cryptography is how similar the math sometimes is. Like, the way that the poseidon hash works (matrix mult -> per-unit nonlinear layer -> matrix mult -> per-unit nonlinear layer) is *exactly* what deep neural networks do.

Can you expand on what this means to you in terms of the interplay between crypto x AI?

steward of @edcon | learned stuff from @variantfund, silverton, first round

There’s a lot of potential for overlap — secure MPC enables multiple dataset owners to combine their sets for private training of a neural network without revealing their inputs to the other parties, yet getting an end output of a network trained on all participants’ data.

BDFL of Quilibrium, Eng @ Farcaster, ex-Coinbase, always opinionated, never hydrated

/quilibrium

This is very interesting and something we are actively researching to enable. Federated Learning essentially reaches the end goal here, but has the potential to leak weights. MPC at the edge can remove that, but where (how?) do you aggregate the weights then? ...

CTO @ Morphic. Previously @ Polygon, @ Uber. Working on disrupting Hollywood with GenAI

https://nf.td/abv

Does using a centralized aggregator (one of the arguments against FL) become available in this case?