80s sci-fi movies gave us rays that miniaturized humans and things without losing a single atom. That's what we call "lossless" compression in computing, where we want to shrink web assets, code, and payloads where 100% bit fidelity is non-negotiable.

As an article that circulated a lot this summer states, compression is prediction. The better a system can predict what comes next, the fewer bits we need to transmit. But we have real engineering constraints!

In this talk, we will explore:

- The classical and modern web stack, how we evolved from basic redundancy to today's landscape
- Why the model must be part of the message
- The format-aware frontier, so how OpenZL uses DAGs, structural knowledge and universal decompressors
- The connection with LLMs, where the prediction of the next token and the cross-entropy loss are mathematically identical to optimal entropy coding... and why language models are great predictors, but impractical compressors