Tokenisation Explained: Why ChatGPT Reads in Chunks #Shorts

Added: 2026-05-06

[00:00:00]Chat GPT doesn't see words, it sees fragments. That famous glitch, count the R's in strawberry. The model doesn't see one word, it sees two objects, straw and berry. The letters inside those blocks are invisible to it. The algorithm is byte pairing coding. Start with raw characters, find the most frequent pair, merge, repeat. Common words survive and unusual ones get shredded into pieces.

[00:00:26]The model's job is just predict the next token, not the next word. This means anything that requires seeing inside a word, such as spelling, is genuinely hard. Language models don't read language, they predict the next chunk of it.