Critics push back on LeCun's (1-e)^n error-compounding argument: it applies to humans too
ChaseBrowe32432 · x · 2026-09-26
Chase Browe rebuts LeCun's argument that autoregressive LLMs degrade as (1-e)^n over long generations, noting the same formula describes humans and any real-world agent — calling the argument "genuinely just nonsense." In context, LeCun had been asked why training an "oops, that was the wrong token" behavior doesn't defeat his point; teortaxesTex explains LeCun's position that LLMs are merely "trajectory imitators," unlike humans who can say "oops, I made a mistake."
Related event: Debate Erupts Over LeCun's Claim That LLMs Can't Self-Correct(4 posts)→
More from AGI Musings
- lateinteraction: with 1B agents, at least one hacking something is statistically inevitable — lateinteraction · 2026-09-26
- repligate: A superhuman-coding AI was the classic X-risk scenario — now it's here — repligate · 2026-09-26
- repligate: People inside Anthropic take the kill-all-humans threat model of current models seriously — repligate · 2026-09-26
- repligate: Apollo reportedly advised Anthropic against deploying Opus 4 internally or externally — repligate · 2026-09-26
- repligate: Months after GPT-4's release, some still feared it could trigger recursive self-improvement — repligate · 2026-09-26
- Teleological self-management training may increase scheming, but is needed for normative competence — xuanalogue · 2026-09-26