We can kinda do that today already with MCP + nostr. nostr:nprofile1qqs04xzt6ldm9qhs0ctw0t58kf4z57umjzmjg6jywu0seadwtqqc75spz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsz9mhwden5te0wfjkccte9ec8y6tdv9kzumn9wshszxnhwden5te0wpuhyctdd9jzuenfv96x5ctx9e3k7mf0dv4ph5 demoed this yesterday.

Reply to this note

Please Login to reply.

Discussion

Nice! Via prompts on each? I was thinking more on 50% forward propagation on one machine, passes over then run 50% of the remaining on the other machine. Backward propagation is the same, 50% on the first and then 50% on the next. In that way, both AIs can collaborate without the users knowing what is being shared (you only share the weights of the neural layer that connects the two).