Newest papers

Predictive Latents for Autoregressive Language Models: Possibility, Gauge, and Obstruction

Yuzhou Hong · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28540 · cited by 0 works (OpenAlex) · open-access PDF

French

Les modèles de langage autorégressifs sont formés par la prochaine entropie des jetons, mais plusieurs systèmes récents répètent également un état caché sans émettre un jeton: pensées continues, étapes de pause et profondeur récurrente. Cette note sépare trois risques sur un processus latent. Le risque de jeton $R_{\mathrm{tok}}$ marque une lecture. Le risque latent forcé par l'enseignant $R_1$ marque une carte d'un pas qui voit l'état caché véritable. Le risque de circuit ouvert $R_H$ marque la même carte après qu'il a été itéré sur ses propres sorties. Si le jeton Bayes suivant est une fonction d'erreur statistique de $k$-dimension avec une covari

Machine translation (Meta M2M100); the English original follows.

Abstract

Autoregressive language models are trained by next-token cross-entropy, yet several recent systems also iterate a hidden state without emitting a token: continuous thoughts, pause steps, and recurrent depth. This note separates three risks on one latent process. The token risk $R_{\mathrm{tok}}$ scores a readout. The teacher-forced latent risk $R_1$ scores a one-step map that sees the true hidden state. The open-loop risk $R_H$ scores the same map after it has been iterated on its own outputs. If the Bayes next-token posterior is a function of a $k$-dimensional statistic with full-rank covariance, a width-$k$ latent, a linear least-squares transition, and a refit unembedding match the Bayes token predictor. The statistic is identified only up to an invertible gauge: whitening the residual stream does not change the token law once the unembedding is refit. The matching obstructions are sharp. A constant latent drives $R_1$ to zero while token cross-entropy falls back to the marginal entropy. A perfect token predictor can still hide a null-space walk whose open-loop error grows linearly in the horizon. A small next-latent loss does not certificate next-token prediction, and a small next-token loss does not certificate free latent rollout, unless a rank or injectivity condition is stated with the number.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Security token, Mathematics, Autoregressive model, Statistic, Statistics, Bayes' theorem

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange