Nathan Lambert pushes back on Dwarkesh: continual learning won't bottleneck AI progress
natolambert · x · 2026-11-14
Nathan Lambert published "Contra Dwarkesh on Continual Learning" on Interconnects, responding to Dwarkesh Patel's widely-read post extending his AI timelines over missing continual learning.
- Lambert argues current systems already amount to AGI; the real question is whether continual learning bottlenecks progress — his answer: it doesn't.
- Dwarkesh's case: LLMs are magical but Fortune 500 firms can't extract humanlike labor because models lack the ability to accumulate experience on the job.
- Lambert's counter: continual learning is the ultimate "algorithmic nerd snipe" for researchers; simply scaling existing systems will yield behavior indistinguishable from human learning — but via a new kind of AI emerging, not a smooth extension of today's LLM stacks.
- He uses the analogy "don't make your airplane too much like a bird" and expects to make some wrong calls while Dwarkesh gets some right, echoing last year's continual learning debate.
A substantive clash between two prominent AI commentators on scaling vs continual learning.
Related event: Debate: Is Continual Learning a Blocker for Automated AI Researchers?(2 posts)→
More from AGI Musings
- AGI endgame: either a balanced multipolar world or a centralized panopticon — CedricMakes · 2026-11-14
- "60% of people around you have some form of AI psychosis," says Alsop III — StewartalsopIII · 2026-11-14
- AI won't make everyone equal—it will split the smart from the rest worse than money — prasenx · 2026-11-14
- Toddlers learn new verbs only from responsive partners; a call to rerun the test with AI — AnnaCiaunica · 2026-11-14
- Developer muses: in Murphy's world, billions of interdependent events make rare failures daily — bingxu_ · 2026-11-14
- Solo dev: agents made me a hundred-decisions-a-day generalist — and unhireable — rudrank · 2026-11-14