AI learns from people. So pay the people.
Every AI model is taught what a good answer looks like by humans. Hugin lets anyone do that work, get paid for it, and help build an AI of our own.
AI is confident, and often wrong
AI assistants answer billions of questions every day. Most answers sound right. Some are not. Here is a real one from Hugin: asked to compare two phones, an AI said the Samsung Galaxy S22 has a 108MP camera.
A REAL HUGIN TASK“It features a 108MP primary sensor…” That is the Galaxy S22 Ultra. The regular S22 has 50MP. The other answer had it right.
A model cannot tell those apart on its own. A person can, in a few seconds. That small judgment, repeated thousands of times, is how models learn to be more correct and more helpful.
The work exists. It just isn't open.
The big AI labs already pay people to rate answers. It happens behind closed doors, through contracts and staffing agencies, with application forms, interviews and identity checks. Most people who could do the work never get the chance.
We think that should be open to anyone with a phone and a wallet.
How Hugin works
- You see a question and two AI answers. Pick the better one: more correct, more helpful, clearer.
- Two other people rate the same pair, without seeing your pick.
- The majority wins. Every pick that matches it earns $0.01 in USDG.
- Withdraw any time once you have $0.50. If you don't, it is paid automatically every night.
New tasks are released every 5 minutes. Each day a fixed amount of the pool is released, and when it is used up, tasks stop until the next day.
No sign-up. No KYC.
You don't need an account, an email address or an ID. You connect a wallet and sign a short message to prove it is yours. That message costs nothing and sends no transaction. Your earnings go to that same wallet.
That's it. You can start rating in under a minute.
Why we pay for answers
- Good data needs real people. Our goal is answers people actually find better, and only people can tell us that.
- Your time is worth something. If your judgment trains an AI, you should share in it.
- Paying for agreement rewards care. You earn when you agree with the majority, so reading properly pays and clicking at random doesn't.
What we use your answers for
Every rated pair becomes a row in a preference dataset: the question, the answer people chose, the one it beat, and how many agreed. We use it to develop our upcoming AI, Hugin, an assistant trained on what real people judged to be the better answer.
The dataset never includes who rated what. Your wallet address is used to pay you, not to describe you.
Fair for everyone
Some pairs are control questions with a known answer, for example a real answer next to one written for a different question. Wallets that miss too many are removed and their balance is not paid. That keeps the pool for people who do the work.
The money is on chain
Rewards sit in a vault contract on Robinhood Chain that anyone can look up. Payouts come straight out of it, in one transaction to everyone who is due, and fees are used to top it up. The vault has a hard daily limit on what can leave it.
The questions and answers you rate are written by open AI models (Llama and Mistral). Rewards are paid in USDG on Robinhood Chain. Nothing on Hugin is financial advice.