how clanker
are you?

// a KL-divergence Turing test, reversed.

Language models predict the next token. You, allegedly, do something more interesting. Finish eight sentences; we measure the divergence between you and the machines.

low divergence = you write like a model. high divergence = congratulations, human.

⚠ demo mode: inference isn't funded yet — scores come from a deterministic stand-in, not the real models.

anonymous · scored only for percentiles · ~2 min

 ▌

3–10 words. be yourself. or don't.

interrogating the models…

token by token. they can't hide their logprobs.

you are clanker

method: for each word you typed, each model reported its top-20 next-token logprobs conditioned on your actual text so far. your words are one-hot distributions, so per-token KL(you ‖ model) = −log p_model(your token), averaged over tokens and questions. words outside the top-20 are floored. similarity is normalized against each model's self-completion baseline, and your overall score is your nearest model's — divergence to an ensemble is a min.