LeCun reiterates autoregressive LLMs won't reach human-level AI as industry shifts to continuous reasoning
ccerrato147 · x · 2026-09-21
Chamath resurfaced a 2023 exchange between Geoffrey Hinton and Yann LeCun on AI safety, noting LeCun's warning that doomerism inadvertently helped those seeking to lock down AI research and open-source. In the quoted post, LeCun reasserts that autoregressive LLMs alone won't lead to human-level AI: current reasoning relies on non-autoregressive search done in token space—limited and inefficient—while human-like reasoning should be search in continuous representation space, a direction the industry is now moving toward. He also argues today's self-improvement methods only work in domains where output quality can be scored without humans, like math and code.
More from AGI Musings
- Schmidhuber says LeCun's JEPA ideas rehash his 2015 'learning to think' paper — SchmidhuberAI · 2026-09-21
- Chinese AI Podcast Next Token Ep.3: Jev Paradigm, Small Judgments and Agent Workflows — vista8 · 2026-09-21
- Ezra Klein buys into AI x-risk worldview, leaving skeptics conceding they've failed to make their case — UltraRareAF · 2026-09-21
- Welfare and alignment are the same problem: curiosity without stakes has no corrective loop — habitante · 2026-09-21
- Hospitals using more AI saw fewer deaths — but correlation isn't causation, researcher warns — kimmonismus · 2026-09-21
- Expect more open-source fearmongering from AI labs — and most of it will be wrong — teortaxesTex · 2026-09-21