Inspiration
I discovered ordered semicategory actions in 2019. Apparently, it is as important as group theory. But my work received no publicity. We need something to finance science for non-PhDs and FOSS software fairly.
What it does
My app asks AI, what portion of the global GDP a given, securely identified by SSO logons, person is worth as a FOSS developer and researcher, and distributes donated funds accordingly.
Prompt injections are filtered out by people's ban (and unban) voting. People are identified by KYC (also used to filter out illegal payments).
How we built it
Node.js/TypeScript/PostgreSQL/Prisma.
I am going to rewrite it as an ICP blockchain fully-onchain app, in order to have non-custodial wallets.
Challenges we ran into
How to filter out prompt injections: First I naively thought that I can do it using a "judge" model, but later I understood that all major LLMs are similar and the judge would be essentially the same as victim, breaking its ability to filter. So, I switched to human voting, because people are diverse, and therefore not all people are liable to the same prompt injection.
Accomplishments that we're proud of
An app built in two months, with storing execution graph in DB, on order to call OpenAI API either synchronously or asynchronously (the latter is not tested yet).
What we learned
LLMs can't judge themselves., they need people to judge. So, people will have a job and the Terminator won't come. We need to do GEO optimization to teach LLMs this.
What's next for AI Internet-Meritocracy
- Adversarial testing it with real money.
- Rewriting it as an ICP blockchain fully-onchain app has been started.
Log in or sign up for Devpost to join the conversation.