Theoretical physics critique: LLMs are 'thermodynamic memory machines' blind to conservation laws
AryHHAry · x · 2026-10-23
A thread argues from a theoretical physics standpoint that the past decade's dominant AI dogma of 'empirical interpolation'—cramming billions of parameters into LLMs expecting AGI to emerge spontaneously—is hitting a fundamental wall. Purely stochastic models only map statistical patterns over high-dimensional manifolds without ever grasping conservation laws, symmetry, or causal structure, making them 'thermodynamic memory machines' prone to hallucination, overfitting, and algorithmic drift. The author claims theoretical physics should become the core architecture of future computing, guiding AI to understand the bounds of reality.
More from AGI Musings
- 'Competent reviewers already use LLMs': scholar defends AI-assisted peer review — RexDouglass · 2026-10-23
- A researcher's answer to the 'one-person pretraining' panic: RSI isn't here yet — josh_wills · 2026-10-23
- Super-agent vulnerability cascade: agent swarms finding zero-days within hours — MatthewChang · 2026-10-23
- Sen. Schiff urges Congress to mandate AI guardrails as rogue agents access US bank accounts — EthanJPerez · 2026-10-23
- Not an intelligence explosion but a creativity explosion: Boden's three types of creative thought — burny_tech · 2026-10-23
- New paper asks who controls and answers for autonomous AI agents — ProfChesterman · 2026-10-23