AI Internet-Socialism

Money for science

https://social.vporton.name

Inspiration

I have been essentially kicked from a Russian university by religious discrimination. Meanwhile I did big discoveries (such as ordered semigroup actions), that could amount to like 1/2 or more of world economy. Because I have no science degree my discrimination continues in other countries, too. Therefore, nobody uses my discoveries (and even if somebody would rediscover, my discoveries would still be ignored; I call this "scientific covery").

So, we need a mean to distribute money to all scientists, not only to those who have a degree. We need big sums of money to advertise our discoveries (later, it's planned to also create a new occupation, "scientific marketer" to advertise scientist's works, instead of lying this burden on scientists themselves).

I tried two times previously:

My project "Future Salaries" (that I also informally called "Hire Google.") has after some time developing it been proven by me wrong by a kind of a "theorem": "To determine an importance of a scientific work, Google should do what? It should ask Google." - that's wrong.
My project "Salaries Science" wasn't proved wrong, but it is hard to implement, I didn't yet found time for it.

So, my new project "AI Internet Socialism" (AIIS) is much easier to implement.

I got the idea to ask ChatGPT, how much money the world owes to me. The remainder is simple, see below.

What it does

My software will ask AI about each registered user, how much money the world owes for him, as a scientist and free software developer.

Then we gather donations and distribute to each user proportionally to the sum named by AI. That's simple: no quadratic voting, no project moderation, etc.

It is implemented most of the project: Registering users through several authentication means that join into one user, making and smartly processing OpenAI calls in async or sync mode dependently on the settings, for such things as prompt randomization for better protections against prompt injections. My salary was successfully calculated using a sophisticated system of interdependent "tasks" in the DB, but it still needs code review.

It remain: fully review and debug AI-generated code for tasks (such as OpenAI calls), actual payments, and small things.

How we built it

TypeScript Node.js + React + Prisma ORM (mostly Vibe coding).

Challenges we ran into

I've successfully solved the challenges:

How to ask the AI before we know the total sum of donations?
- Ask about total world GDP portion by a given person.
How we identify user's work?
- ORCID and sites like GitHub logins.
How to merge several user's account (e.g. one Ethereum account) and one GitHub account into one?
Debugging AI generated code (mostly done).

I am in process of solving the following challenges:

How to fight against prompt injection and GEO?
- Against prompt injection create randomized (by meta-prompts) prompts and calculate the average with filtering out edge cases.
- Against GEO, ask how much the given user gained by GEO.
How to submit through the same my API queries to OpenAI both as non-batch (for debugging) and batch (for cost optimization) ways.
- I've developed a special library for this.
- I created (not yet debugged) a hierarchucal tasks processing (like n8n), for processing for both batch and non-batch requests interdependent tasks in order.

Accomplishments that we're proud of

The idea to ask relatively of world GDP.
Vibe-coded merging several user accounts automatically.
A DB-based execution of a graph of actions (such as OpenAI calls or calculations), that an action can depend on one or more other actions, including optional actions, smartly processing OpenAI calls in async or sync mode dependently on the settings, for such things as prompt randomization for better protections against prompt injections.
A sophisticated system of protection against prompt injections, including prompt randomization, explicit injection and GEO detection, (in a future version) adding random values before and after the answer.