Perplexed
How Perplexed Works
You are the language model now. We show you a sentence with the last word amputated. Your one job — the only job you were ever trained for — is to predict that next token. We already asked a model and froze its probability distribution. Match it.
- Read the sentence prefix.
- Type your guess for the next word (suggestions appear as you type), or flip on Easy mode to pick from a shortlist.
- Hit Predict. You score the probability mass the model assigned to your exact word.
- The full distribution is revealed. Bask in your loss. Repeat for ~11 rounds.
How Scoring Works
Points are the model's confidence in your token. The single most likely word pays the most; a plausible alternate pays partial credit; an off-distribution hallucination pays exactly zero and quietly raises your perplexity. Your accuracy is the share of the total available mass you captured.
Daily Run
Tap Daily run for a deterministic set of prompts seeded from today's date — everyone on Earth gets the same sentences, so you can argue about them. Keep a streak going by clearing the daily every day.
Slop Fact: Perplexity is just the exponential of the average surprise per token. A perfect model has perplexity 1 and never learns anything new. You, gloriously, have a perplexity in the dozens — which is to say you are still capable of being surprised, unlike the thing that wrote this sentence.