Onchain AI - Favourite Conversations ❤️

I’ve only recently started paying attention to DeAI on the IC. I’ve known there have been onchain models for a little while but didn’t really know where to access or interact with them.

I think this represents a USP for the IC, and is really a space where more usage and innovative will help elevate the entire IC ecosystem.

To that end I thought I’d start a thread to encourage people to check out IConfucius (an LLM running fully on-chain), and a community-produced frontend (via Caffeine AI) through which you can chat to IConfucius. My hat goes off to the ONICAI team and their passionate community.

Interact :technologist:

Build :construction_worker:

Explore :eyes:

:index_pointing_up:


In the meantime, lets all start chatting with IConfucius and observe the network burn rate spiking. Share your favourite interactions below! I’ll get the ball rolling →

Here’s another onchain model in case IConfucius gets overloaded. Again, hats off to ONICAI.


Early days, so be prepared for hallucinations and performance limitations.

:rofl:

Can definitely be done with the on-chain Qwen 2.5.

Bootstrapping the prompt a bit using icgpt.onicai.com

Once the prompt is designed, this can be coded up in a dedicated LLM, and voila, there is your game!

The prompt can be optimized, such that it does not need to re-ingest it all the time, and the response will come back much quicker.

The prompt can then also be parameterized, so you can use {word} instead of London…

Thanks @icpp, would ONICAI consider exposing a service like this, with all of the optimisations you have described? A caller provides a word, and the service provides a clue. There are typically billions of words in English corpora, and many more if you allow for other languages, so pre-storing many clues for all words in a canister isn’t feasible. Generating them on the fly requires an on-chain LLM if we want verifiability (that the caller and service are behaving as prescribed for all users). Also it’s a nice on-chain LLM demonstration with an end-user-facing function.

Imagine a game like Wordle (hugely popular) on the Internet Computer, but harder game play (variable length words) and with the ability to ask an on-chain AI for clues. There could be a very small pay-to-play or pay-to-guess, with a daily payout for the fastest user and also the user who guessed with the fewest tries and/or clues (you’d need each user to get a different word each day to avoid gaming the system via sybil accounts).

The service provided by ONICAI could be very specific in scope (give a word, get a clue). There are a huge variety of games that consuming services could produce with this. The one described above, crossword puzzle games, or scrabble-like games, or others.

Users love to play games like these, and if they can earn money for playing well, while experiencing something that can only exist on the IC, facilitated by ONICAI, that would be awesome!

@WebTreeSoftwareSolut, I bet with your Caffeine AI prowess you could build a consumer of a service like that and package it in a game. An AI-generated game consuming an on-chain AI service. That would really showcase what the IC is about.

@dominicwilliams, @Lomesh-dfn1, @ld-dfn1, I hope you’re gearing up to show support for ONICAI in their upcoming SNS decentralisation sale.

@lorimer,

I guess we’ll never know.

Anyways, good luck with your SNS guys, you seem like the real deal to me, you have my endorsement. @icpp

Absolutely.
If the onicai SNS is successful, this can be scoped as a development project, with milestones & deliverables. And the community helps define it and gets the final vote if it makes sense to create it.