Kids are effectively 10M-parameter LLMs running at temperature 2.0

bronzeagepapi · x · 2026-09-29

A widely shared quip makes the case that children are basically tiny language models: 10M parameters of knowledge, but sampled at temperature 2.0 — highly stochastic outputs, wild tangents, and occasionally brilliant surprises.

It's a funny framing that uses LLM vocabulary to capture exactly why kids say such unpredictable things.

Original post →

More from Fun

Fun channel →