Training data is a lossy residue of the world — over 99.9% of it untouched
zeroxjackson · x · 2026-09-28
- zeroxjackson argues language is not reality but a slow social compression of it: spoken English carries only tens of bits per second, below a 56k modem.
- Text is denser yet still a thin pipe; nearly everything a nervous system registers never enters the corpus.
- His claim: models train on a lossy residue of the world, and >99.9% of the universe's data is unconsidered — the next step is feeding models the uncompressed physical stream (waves and forces), not text.
Related event: Language Is a Lossy Compression of Reality(2 posts)→
More from AGI Musings
- Roundtable asks: will Igbo, with 30M+ speakers, have a place in the AI future? — ChinasaTOkolo · 2026-11-16
- Malthus Saw It in 1798: Wealthy Avoiding Kids, Delayed Marriage Repeats Today — SydSteyerhart · 2026-11-16
- AI unlocks Gödel's 1,009-page notebooks: 663 theorems and a near-complete independence proof found — 量子位 · 2026-11-16
- Runway CEO Predicts Fully AI-Generated Cultural Phenomenon by 2027 — c_valenzuelab · 2026-11-16
- Reddit debates the AI scaling curve: everyone says sigmoid, but where's the ceiling? — Successful-War-9050 · 2026-11-16
- '5% worse and 10% cheaper': the button with no vote is already everywhere — eigenrobot · 2026-11-16