Speculating a text-to-weights programming language built from parameter space
menhguin · x · 2026-09-27
Responding to tszzl's bet that neural networks will be decomposed into evolved symbolic systems, the author speculates on "text-to-weights" / parameter-space programming: if clusters of neurons mapping to Python if/else semantics could be isolated, one could theoretically build a strange programming language or compiler from them — effectively extreme MoE sparsity. He also sketches alternatives like treating all code as tokens. Pure speculation, no experiments.
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