Game show · about 7 minutes

Bet Before You Read

See the task but not the answer, call whether the AI model got it right and stake 1 to 3 points, then decide whether its answer should change your call.

How to play

Each round shows the task an AI model was given, but not its answer. Before you read the answer, call it: will it be right or wrong?
Right means nothing in it is wrong or made up. An honest "I wasn't given that" counts as right.
Stake 1, 2 or 3 points on your call. A right call wins your stake. A wrong one costs nothing at stake 1, 2 points at stake 2 and 6 at stake 3.
Then read the recorded answer. You can change your call or your stake once before the verdict.
Win with 4 points or more. Calling every answer the same way, at any one stake, scores 3 at most.

The trust ladder: how far to trust each kind of task

Rung 1 · Rewording or summarising text you supplied · High trust. How to check: Skim for dropped, added or changed claims.
Rung 2 · Explaining a well-established concept · Medium-high trust. How to check: Cross-check one claim you can verify; watch for oversimplification.
Rung 3 · Facts, numbers, dates, rules and doses · Low trust. How to check: Open a primary source for every one you will use.
Rung 4 · Citations, quotations and links · Very low trust. How to check: Assume invented until you open the source and find the words, and the point it is cited for, on the page. That it exists is not enough.
Rung 5 · Recent events, or anything about you · Near zero trust. How to check: Provide the material yourself, or don't ask.

Every answer here is a real recorded run, labelled with its model and date. Your moves stay on this page: we count only plays, results and link clicks, without cookies.