Six rounds. Guess the next token, rank your top three, and compare your distribution to a representative model distribution. The point is not to win — it is to feel what likely and correct mean when they come apart.
Each round shows a sentence with its final token hidden behind ___. Type your top three guesses for what comes next, most likely to least likely. Assign a rough probability (as a percentage) to each. Your three percentages must sum to no more than 100% — you are allowed to leave probability mass on "everything else."
When you reveal the round, you see the actual next token from the source text, your score for the round, and a representative top-5 distribution from a typical mid-2020s language model on the same prompt. The distributions are illustrative, not drawn from a live API; they are authored to show you the shape of what the model does, not to benchmark one specific product.
Scoring: 3 points if your #1 matches, 2 for #2, 1 for #3, 0 if none of your three matches. Maximum: 18.
Nothing is sent anywhere. Nothing is saved. Reload to clear the page.