I think it is necessary to compare the estimated costs also with big tech clouds, for each of the possible solutions even if in an approximate figure. If the project really aims to compete with the current cloud.
The ICP website and forum are now powered by askAI. This is a web2 kappa AI that trains the ICP documentation and
It should be possible to replace this with 8B Llama and RAG, all running in Canister on ICP. Unlike learning, I’ve heard that the output from RAG can vary in performance, but it’s a great object for a challenge. The objectives of this demo can be considered as follows:
-
To run a full on-chain AI on the ICP and establish a position in DeAI.
-
Kinic’s vector DB is competitive enough compared to other clouds to win customers.
-
To anticipate increased use of the system for documentation of other chains, etc.
If askAI can be completed on ICP, it will pave the way for practical operation on ICP, not only for web3 but also for web2. It would also pave the way for a new AI field of trustless RAG.Could you please consider this?
3 Likes
i finally got a chance to publish it: Is AI on smart contracts actually here? | by Diego Prats | The Internet Computer Review | Oct, 2024 | Medium
5 Likes
Hey @diegop, can you please push for improvements to http outcalls so we can have verifiable AI at least through them. For example: HTTP outcalls: HTTP request error. RejectionCode: SysFatal, Error: Timeout expired - #12 by Mar
1 Like
Let me escalate it! To make sure people see it
1 Like