Are You Just Autocomplete?

If an AI can guess your next word, how much of you is autocomplete? I gave a small AI 322 real comments people wrote online and asked it to guess every next word. It got 28% right. But look at which words: the glue words ("the", "to", "of") 40% of the time, the words that carry meaning only 19%. And neuroscience says your brain is a prediction machine too. 0:00 Finish this sentence 0:38 "It only predicts the next word" 1:27 Your brain guesses too 2:33 How predictable are you? 3:29 Which words it could guess 4:24 The face you didn't pick 5:21 Shannon: where the message lives 6:05 So, are you autocomplete? METHOD: Qwen3-8B (4-bit, MLX), run on my own computer. 322 public Hacker News comments written 25 Oct 2026 (after the AI's training; plain prose, 40–150 words, no code or links). At every word after the first five, I took the AI's single top guess for the first chunk of the next word: 22,163 words scored. No usernames or comment text are shown. RESULTS: top guess right 28.4%; right word in its top 5: 53.1%; glue words 40.2%; meaning words 19.2%; 37% of words got under a 1% chance. "I'll have a cup of": tea 60%, coffee 25%, soup 0.1%. The zenith line: the AI predicted "me", "is", "in", "the" and "couldn't"; its top guess for "part" was "unpredictable". CREDIT: the question comes from Scott Alexander's essay below and the "next-token predictor" debate on Hacker News (Sep 2026). SOURCES: Kutas & Hillyard (1980), Science 207:203–205 (the N400); example sentence from Kutas & Federmeier (2011), Annual Review of Psychology Clark (2013), "Whatever next? Predictive brains, situated agents, and the future of cognitive science", Behavioral and Brain Sciences Schrimpf et al. (2021), PNAS: https://pmc.ncbi.nlm.nih.gov/articles... Goldstein et al. (2022), Nature Neuroscience: https://pmc.ncbi.nlm.nih.gov/articles... Caucheteux, Gramfort & King (2023), Nature Human Behaviour: https://pmc.ncbi.nlm.nih.gov/articles... Shannon (1951), "Prediction and Entropy of Printed English", Bell System Technical Journal Bender, Gebru et al. (2021), "On the Dangers of Stochastic Parrots", FAccT Geoffrey Hinton, CBS 60 Minutes (Oct 2023) Scott Alexander, "Next-Token Predictor Is An AI's Job, Not Its Species" (Feb 2026): https://www.astralcodexten.com/p/next... Anil Seth, TED 2017, "Your brain hallucinates your conscious reality" Johansson, Hall, Sikström & Olsson (2005), "Failure to detect mismatches between intention and outcome in a simple decision task", Science 310:116–119 (choice blindness) Erman & Warren (2000), "The idiom principle and the open choice principle", Text 20(1):29–62 Anthropic (2025), "Tracing the thoughts of a large language model": https://www.anthropic.com/research/tr... Geoffrey Hinton, Nobel Prize in Physics 2024 (shared with John Hopfield) Narration: AI voice (one designed narrator voice, open-source model run locally). Made with the help of AI tools, and checked by a human. Found an error? Say so in the comments and I'll pin a correction. What's the least predictable thing you've ever said?