Can current LLM architecture reach AGI? An engineer lays out his doubts
mostly_deterministic · reddit · 2026-09-08
A software engineer with 20+ years of experience posted a long Reddit question on whether current LLM architecture can achieve AGI. His core argument: autoregressive next-token prediction gives the model no independent mechanism to know when it's wrong; inference doesn't update weights; generation can't backtrack and revise tokens; agent harnesses add what looks like intelligence but are just layered probabilistic systems — like a room of clones playing near-lossless telephone. Humans can backtrack and refine thoughts on their own, while the model's primitive is generation, not explicit reasoning. He asks whether AGI claims on this architecture are smoke and mirrors or whether he's missing something.
More from AGI Musings
- KAN author Ziming Liu warns AutoResearch generates chaos, not insight — ZimingLiu11 · 2026-10-27
- All 8 Transformer paper authors left Google, and rival labs captured most of the frontier value — signulll · 2026-10-27
- Anthropic's S-1 pledges AI for curing diseases and serving 4.6B underserved patients — petergyang · 2026-10-27
- Goldman finds statistically significant negative link between AI adoption and US employment growth — ben_j_todd · 2026-10-27
- a16z: World-changing tech makes a chaotic entrance — AI agents are following the same path — a16z · 2026-10-27
- NEJM AI on medical education: mastery requires performing without the machine — BraydonDymm · 2026-10-27