Federated Learning with PoW is the only way to "Decentralized AI"
Public.icon
Federated Learning based on PoW is the only way to "Decentralized AI"\nfrom Federated Learning
it's the only way to come to true "Decentralized AI"Previous companies, interviews and appearances related to oneself.icon
if A future where each DAO has AI, maybe choice this structure.
How trade each model?
based on incentive
Federated Learning with PoW
it's let's grow this model for each member in this DAO
some user provide model in FL
some user work Semantic Annotation
The way everyone is donating data to grow the machine learning model. That will look like paying taxes. painted by DALL·E
https://gyazo.com/cee481f04e463c7ae8b79e389b1277ae
Tweet from shogochiai.icon, maybe something similar to what is being imagined?
@shogochiai: @itaihito2 I've been worried about the structure that becomes dependent on politics as long as Chatbots trained by large-scale language models are provided via HTTP APIs, and I've been thinking recently that I want something like being able to put a trained model on the terminal side.
Issues with censored pre-trained models for ChatGPT
@shogochiai: However, it seems difficult to do feedback learning when placing the trained model on the terminal side.
@shogochiai: Rather than having the terminal side hold the learning model, I thought that, if we consider the server side as blockchain-like and make all nodes do the same feedback learning, we could access it without any problem even if one of the nodes were censored.
@itaihito2: @shogochiai I'm still not sure how to do it, but the combination of POW and AI censorship resistance is very interesting.
>Rather than having the terminal side hold the learning model, I thought that, if we consider the server side as blockchain-like and make all nodes do the same feedback learning, we could access it without any problem even if one of the nodes were censored.