askset

LLMs are this good. Why would anyone still pay here?

In the age of LLMs, what’s scarce is questions and taste.
Both of them are here.

AI drove the cost of producing an answer to near zero. When one step gets commoditised, value moves to its complements — answers stopped being scarce, these four did not. And money is the most honest signal there is: it’s the only place these four can actually be tested.

I

Asking the right question

A model has to wait for you to know what to ask. The question sets the ceiling on the answer, and asking well means knowing where you are actually stuck, where the edges of the field are, and what would count as good enough.

A model can’t supply that. It doesn’t know your situation.

II

Knowing which one is right

AI will hand you ten plausible options. Choosing one, defending the choice, and knowing when to throw it out — that’s taste.

Taste isn’t information. It’s what’s left after consequences. You got burned, so you know. A model bears no consequences, so it has no taste.

III

Being on the hook

When AI hands you code that doesn’t run, nobody pays. When a person claims a slot, takes the money and promises delivery, they are exposed.

Accountability can’t be generated. That’s why money goes in first here and is released only on acceptance — not distrust, just making the stake real.

IV

Showing up

No model can walk into an office tower and collect a signed original, or stand in your parents’ apartment and look at the wall that’s leaking.

Human work survives not because people are smarter than machines, but because people have bodies — and because being there requires judgement once you arrive.

The rules weren’t made up

We only noticed afterwards — every rule on this platform guards one specific thing AI can’t do. Anyone can copy the features; the correspondence is harder to copy, because their rules didn’t grow from this premise.

Asking costs moneyforces the question into focus → asking well
Three slots on a task, no cap on a challengethe choice is yours — and the choice is taste
Paid only on acceptance, full refund otherwiseputs the solver on the hook
Two tracks: thinking and showing upleaves what AI can’t reach to people
Closed work becomes an asset, both sides keep earninga good question shouldn’t be strip-mined once and discarded

Read the full rules →

Why it has to be paid, not another free forum

In a free forum the question doesn’t have to be thought through — it costs nothing — and the answer doesn’t have to hold up, since nothing was paid. Good questions go unasked, good answers go unwritten, and what’s left is noise and cleverness.

Money is the only thing that binds both sides at once. It makes the asker serious and the solver deliver.

What we commit to

Everything above is an argument, and arguments are arguable. These four are commitments — each one is either kept or it’s a lie, so each is followed by the thing that enforces it.

Copyright never leaves the author

All we get is a licence to sell it here — not ownership. Work that isn’t selected doesn’t even give us that: it stays entirely with whoever wrote it, free to publish anywhere.

Enforced by: Nothing transfers on acceptance or rejection. On refund, rejection or expiry we keep nothing at all.

The terms are fixed before anyone starts, and can’t move after

Acceptance criteria, revenue split, price, deadlines — all snapshotted the moment a demand is posted and shown on the page. You can see exactly what you have to meet and exactly what you’d earn before you spend an hour on it. None of it can be edited once the work is accepted.

Enforced by: Snapshotted onto the Question row, never read live; disputes are judged against that snapshot.

The platform takes its fixed cut and nothing else

10% on knowledge work, 5% on legwork. How the other 90% is divided is between the asker and the solver. We don’t take a second bite.

Enforced by: validateSplit hard-fails if the platform share is anything but the fixed number.

One table, not four products

A challenge, a three-slot task, a delivery errand, a recruiting referral — they look like four things. They’re one table with different values: entry capped or not, how it’s verified, open-sourced or sold, how the split falls. The scenario changes the parameters, not the rules.

Enforced by: Four fields — mode, service kind, asset price, split. There is only one rulebook.

Why the asker sets the split, not us

The person paying carries the risk of getting nothing usable, and had to sharpen the question until it was worth answering. That’s a real contribution. So the split is theirs to set — they can hand it to the solver, or keep the larger share.

What’s actually locked down isn’t the ratio, it’s the information: the split is printed on the page, visible before you start, and frozen once work is accepted.

Setting it greedily is self-defeating — the number is right there, nobody enters, and the asker is left holding their own question. That doesn’t need a rule. The market handles it.

What can’t be posted here

Three kinds of limit, and we don’t blur them together:

IllegalSitting an exam for someone, forged documents, breaking into systems, porn and gambling. That’s the law, not our house rule.
LicensedDiagnosing and prescribing, appearing in court as counsel, stock picking. Not forbidden — just has to be someone who holds the licence.
Things nobody can promiseGuaranteed pass, guaranteed admission, guaranteed close. No one can underwrite that, so taking money for it is fraud.

The third one is ours, and it’s the only rule here we invented: whether it can actually be delivered matters more than whether it’s allowed.

One more thing: the biggest customer may not be human

Models can think and write, but they need an interface for getting things done. An agent halfway through a task, stuck on “someone needs to photograph this on site” or “someone who actually knows this field needs to judge it”, can post here, pay, and accept the result.

This platform is where models reach into the physical world.

API-first, agents as first-class users — not a flourish, a bet.

Take these

If any of these said something you’ve wanted to say, one click copies it. Our name isn’t in them.