WeSearch

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

·3 min read · 0 reactions · 0 comments · 6 views
#beyond#tracking#shortcut#composition-bounded#predictive
Beyond Tracking or Shortcut: Composition-Bounded Predictive States in Poker Autoregressive Models
TL;DR · WeSearch summary

This paper studies this ambiguity in a no-range Limit Hold'em autoregressive model trained only on action and value targets, not on an opponent's hand or range. Opponent-range probes are positive after action/value controls in two of three seeds, and the behavior head predicts held-out actions about five percentage points above a baseline using only observable public history. However, visible public betting composition explains more opponent-range signal than residual hidden states, suggesting that most recoverable information comes from betting summaries.

Key facts
Original article
arXiv.org
Read full at arXiv.org →
Opening excerpt (first ~120 words) tap to expand

Computer Science > Artificial Intelligence arXiv:2607.19369 (cs) [Submitted on 13 Jun 2026] Title:Beyond Tracking or Shortcut: Composition-Bounded Predictive States in Poker Autoregressive Models Authors:Quanhao Li, Qianyu Chen View a PDF of the paper titled Beyond Tracking or Shortcut: Composition-Bounded Predictive States in Poker Autoregressive Models, by Quanhao Li and 1 other authors View PDF HTML (experimental) Abstract: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 over hidden states. This paper studies this ambiguity in a no-range Limit Hold'em autoregressive model trained only on action and value targets, not on an opponent's hand or range.

Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv.org.

Anonymous · no account needed
Share 𝕏 Facebook Reddit LinkedIn Threads WhatsApp Bluesky Mastodon Email

Discussion

0 comments

More from arXiv.org