Grady Booch: transformer LLMs remain non-deterministic statistical engines at their core

Grady_Booch · x · 2026-09-22

Grady Booch argues that the architecture of every transformer-based LLM yields next-token prediction. Symbolic harnesses surrounding an LLM soften those sharp edges, but at the core they remain non-deterministic, statistical engines walking through a large static multi-dimensional manifold.

Original post →

More from AGI Musings

AGI Musings channel →