works on
attention dynamics 1efficiency 1multimodal large language models 1token pruning 1training-free methods 1

From the 1 of 2 linked papers with an AI index.

collaborators
Showing cs.AIShow all

1 paper · 1 filter

cs.AI2026

Beyond Tracking or Shortcut: Composition-Bounded Predictive States in Poker Autoregressive Models

Quanhao Li, Qianyu Chen

Hidden-state probes often recover latent labels in imperfect-information sequence models, but this alone does not establish that a model maintains a posterior belief distribution o…