---

What happens when you treat language models like noisy radio signals?

---

Imagine trying to listen to a radio station in a thunderstorm — the more you crank up the volume (scale), the more static creeps in. A new theory borrowed from 1940s information theory explains why bigger isn't always better for AI training.

This means researchers can now predict *when* scaling will help and when it'll actually hurt — like catastrophic overtraining — instead of just hoping bigger models work.

---