The paper investigates whether deeper predictor blocks in adaptive‑compute world models truly improve rollout accuracy and whether a router can select the appropriate depth per step. Experiments on nine DeepMind Control tasks show that deeper predictors do not always yield better accuracy, and that a greedy oracle selecting depth per step can reduce error by up to 37% compared to fixed‑depth mixtures, yet no trained router achieves this advantage. The study also finds that supervising early exits only at the first rollout step can mitigate depth‑inversion issues observed in some tasks.
arXiv:2607. 10203v1 Announce Type: cross Abstract: Adaptive-compute world models -- early-exit or mixture-of-depths predictors that spend variable depth per step -- assume depth buys better predictions and can be routed adaptively.
arXiv:2607. 10203v2 Announce Type: replace-cross Abstract: Adaptive-compute world models -- early-exit or mixture-of-depths predictors that spend variable depth per step -- assume depth buys better predictions and can be routed adaptively.
arXiv:2610.15218v1 Announce Type: new Abstract: A recurrent-depth language model reuses one block of weights several times in a row. The number of repetitions, which we call loops, is chosen when the...
arXiv:2608. 11233v1 Announce Type: cross Abstract: A dense, pretrained language model can be retrofitted with recurrent depth and learn an iterative latent transition that persists after outcome-only annealing.
arXiv:2606. 01838v1 Announce Type: cross Abstract: Agentic language model systems alternate between two structurally distinct step types: structured tool calls (short, deterministic, low perplexity) and open-ended planning/reasoning steps (long, complex, high perplexity).
arXiv:2607. 14427v1 Announce Type: new Abstract: A depth-recurrent transformer applies a weight-tied core a variable number of times, and prior work has shown that training with a randomized recursion count yields one checkpoint usable across a range of inference depths.
The paper challenges the common practice of selecting latent world models based on one‑step prediction loss, arguing that this metric poorly predicts planning performance. Using a frozen‑decoder protocol, the authors decompose one‑step error into encoder‑inherent and predictor‑added components and evaluate goal‑conditioned planning across three continuous‑control environments. Their experiments show that improvements in one‑step fidelity do not correlate with planning success, with correlation coefficients near zero and inconsistent exchange rates across tasks.
arXiv:2603.21676v2 Announce Type: replace-cross Abstract: Standard Transformers have a fixed computational depth, limiting their ability to generalize to tasks that require variable-depth reasoning....
The paper introduces SLIM, a pushing benchmark that tests language‑based goal specification for latent world models. It shows that a vanilla LeWM fails almost entirely due to an action‑insensitive encoder, but adding an inverse‑dynamics auxiliary loss restores encoder sensitivity and dramatically improves success rates. With the repaired latent, a lightweight language‑goal head can plan from natural language sentences, achieving high performance on navigation and pushing tasks without retraining the world model.
The paper introduces a selection‑safe conformal early‑exit method that calibrates a single split‑conformal quantile using a terminal‑path score derived from prefix maxima. This approach guarantees coverage for any data‑dependent stopping policy that preserves terminal‑set containment, without requiring an α/L correction. Empirical results on eight encoder backbones and CIFAR‑100 show that pathwise selection achieves higher macro coverage (0.901 vs. 0.814) with smaller set sizes and lower latency fractions compared to naive layer‑wise selection, and remains robust to deployment‑time budget changes.
The paper investigates how the dynamical regime of recurrent-depth reasoners—whether they settle, drift, or remain marginal—affects the reliability of test‑time depth. It establishes a depth‑safety condition based on per‑step displacement relative to the decoder margin, showing that operators in a settling regime can safely increase depth without degrading performance and can even improve accuracy on harder unseen tasks such as Sudoku. The authors provide empirical evidence from algorithmic tasks trained on limited data, demonstrate the impact of a terminal fixed‑point objective on depth behavior, and offer operational criteria to identify useful test‑time depth while cataloguing failure modes.
LEV is a decision model that performs reasoning by looping its layers instead of generating text, allowing it to deepen its computation on a per-question basis. Trained jointly across all loop steps, a single LEV checkpoint can answer at any depth, improving accuracy—especially on hard questions—while keeping inference fast. On benchmark tests, LEV outperforms dense and cascade models by 5.4 to 8.8 accuracy points with sub‑50 ms latency, and on‑demand depth selection saves 15–24% of computation for minimal accuracy loss.