Study Highlights Utility‑Learning Tension in Self‑Modifying AI Agents

The paper, submitted 5 Oct 2025, shows that unrestrained utility‑maximising changes quickly degrade learnability, while a two‑gate policy preserves model capacity. getnews.me/study-highlights-utility... #ai #selfmodifying #aisafety

SMITH An assembly-like language with no jumps or branches whatsoever.

Cat's Eye Technologies
SMITH
2000

A language with no branches or jumps can still be Turing-complete if its programs can modify themselves indefinitely.

codeberg.org/catseye/SMIT... #esolang #esolangs #programminglanguages #selfmodifying