Hard state tracking may be a sample-complexity problem, not data scarcity
lambdaviking · x · 2026-10-04
A technical X thread on why models struggle with hard state tracking: rather than such patterns being rare in natural pretraining data, lambdaviking suggests they may simply have higher sample complexity under gradient descent compared to simpler kinds of state tracking — with the caveat that this is speculative territory.
More from Research
- Living Science proposal: papers as living, updatable units of scientific knowledge — ChenhaoTan · 2026-10-05
- Author seeks community review of computer vision textbook draft converted from notes with Claude — baadshaha · 2026-10-05
- Cohere Labs to host talk on Building Trustworthy AI for Mathematics — Cohere_Labs · 2026-10-05
- ICML 2026 proceedings go live on PMLR as Volume 306 from Seoul conference — lawrennd · 2026-10-05
- Neuroscientist Anil Seth's TED talk: current AI won't become conscious — we're seeing faces in clouds — anilkseth · 2026-10-05
- CMU ships cua-speedrun: standardized benchmark finally measures computer-use agent speed and cost — rsalakhu · 2026-10-05