August 11, 2026
Zip it... or predict it?
Compression Is Prediction
Turns out chatbots and file-zippers may be playing the same guessing game — and the comments erupted
TLDR: The article argues that shrinking data and training chatbots are basically the same game: getting good at prediction. Commenters loved the brain-bending idea, but some pushed back hard, saying the author oversold chatbots and ignored the bigger truth that many kinds of prediction systems work this way.
A blog post trying to explain why data compression and AI chatbots are secretly doing the same basic trick somehow turned into a full-blown comments-section philosophy fight. The article’s big idea is simple enough for non-engineers: whether you’re shrinking a file or training a chatbot, the system is trying to get really good at guessing what comes next. If it can predict well, it can store or generate information more efficiently. Mind-blown? Yes. Universally accepted? Absolutely not.
The crowd split fast. One camp was thrilled, with commenters saying this helps answer the sneering old line that chatbots are “just next-word predictors.” Their comeback: if prediction is really a form of compression, then maybe “just predicting” is a lot more powerful than critics want to admit. Another group immediately went into well, actually mode, arguing the article aimed too narrowly at large language models and should have said all statistical models are compressors. One commenter even dragged in embeddings and auto-encoders like an academic receipt drop.
There was also some classic internet comedy in the mix: one user showed up only to announce they’d rushed to post and discovered someone had already beaten them to it — a tiny tragedy of online life. Others tossed in homework links, including a Grant Sanderson video and older neural-network compression threads, turning the discussion into equal parts mind-expanding, nitpicky, and peak comment-section chaos.
Key Points
- •The article distinguishes minification from compression, stating that minification removes nonessential syntax for machines while compression relies on redundancy in data.
- •A JavaScript example is used to show minification reducing code from 156 characters to 62 characters.
- •The article explains redundancy-based compression with a repeated-character string and demonstrates run-length encoding.
- •Using 8-bit ASCII, the article compares the repeated string's size before and after run-length encoding: 224 bits versus 96 bits.
- •The article describes modern compression tools as consisting of transforms, models, and entropy coders, with models supplying symbol probabilities to entropy coders.