One-Step Next-Latent Prediction Is Not a World Model
Abstract
Next-latent prediction fits a map from the current embedding to the next one. LeNEPA carries this objective to time series, replacing the stop-gradient of next-embedding prediction with the isotropy penalty of LeJEPA. A world model is a transition kernel that can be rolled out. The one-step regression identifies a conditional mean, and a mean is a kernel only in special cases. For a linear-Gaussian Markov latent, the mean transition and the innovation covariance are fixed by the one-step problem, and the open-loop squared error at horizon K equals the trace of the sum of the pushed-forward innovation covariances. That error grows with K after the one-step fit is exact. If the conditional mean is nonlinear, composing it is not the multi-step conditional mean. If the observation is a non-injective function of a Markov state, a memoryless one-step map does not determine future observations, while a short window can. An isotropy penalty is a function of the embedding marginal, so its partial derivative in the transition weights is zero. On a scalar autoregression with coefficient $0.9$, the one-step mean squared error is $0.998$ and the $16$-step open-loop error is $5.10$. On a hidden rotation, an eight-step window reaches $16$-step error $0.056$, while the current scalar alone reaches $0.778$. Raising the isotropy weight from $0.1$ to $10$ leaves eight-step latent error inside $[0.78,0.85]$ on three seeds.
Related
- Learning Stabilizing Latent Feedback Controllers via Multi-Step Backward Losses
- What Comes Next and Why: Interpretable Next-Event Prediction with Neuro-Symbolic Rules
- Density-Guided Bootstrap Boundaries for Latent World-Model Reinforcement Learning
- Information-Theoretic Analysis of Next-Token Prediction under Markovian Data
- Pooling interval-monitored step-stress accelerated life tests for reliability prediction: A failure-rate-based meta-analysis for the proportional hazards family
BibTeX
@misc{wang2026onestep,
title = {{One-Step Next-Latent Prediction Is Not a World Model}},
author = {Shitong Wang and Zhongang Cai and Yuzhou Hong},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.36227},
eprint = {2609.36227},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.36227},
}
RIS
TY - UNPB
TI - One-Step Next-Latent Prediction Is Not a World Model
AU - Shitong Wang
AU - Zhongang Cai
AU - Yuzhou Hong
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2609.36227
UR - https://arxiv.org/abs/2609.36227
AB - Next-latent prediction fits a map from the current embedding to the next one. LeNEPA carries this objective to time series, replacing the stop-gradient of next-embedding prediction with the isotropy penalty of LeJEPA. A world model is a transition kernel that can be rolled out. The one-step regression identifies a conditional mean, and a mean is a kernel only in special cases. For a linear-Gaussian Markov latent, the mean transition and the innovation covariance are fixed by the one-step problem, and the open-loop squared error at horizon K equals the trace of the sum of the pushed-forward innovation covariances. That error grows with K after the one-step fit is exact. If the conditional mean is nonlinear, composing it is not the multi-step conditional mean. If the observation is a non-injective function of a Markov state, a memoryless one-step map does not determine future observations, while a short window can. An isotropy penalty is a function of the embedding marginal, so its partial derivative in the transition weights is zero. On a scalar autoregression with coefficient $0.9$, the one-step mean squared error is $0.998$ and the $16$-step open-loop error is $5.10$. On a hidden rotation, an eight-step window reaches $16$-step error $0.056$, while the current scalar alone reaches $0.778$. Raising the isotropy weight from $0.1$ to $10$ leaves eight-step latent error inside $[0.78,0.85]$ on three seeds.
ER -