arXiv Computation and Language October 1, 2026 By Xiaoqiang Wang, Suyuchen Wang, Bang Liu

LatentHarness: Learning Latent Actions for Memory and Reasoning via Counterfactual Policy Distillation

Read the original on arXiv Computation and Language →

LatentHarness unifies memory access and latent reasoning by treating them as sequential latent actions—THINK, RECALL, and EXIT—within a language model. It is trained via counterfactual policy distillation, which evaluates the impact of each action on the emitted token and learns when to recall evidence versus continue reasoning. On six long‑context reasoning benchmarks, a 1.4B‑parameter LatentHarness model outperforms the strongest baselines by 2.8% and 10.0% relative, while running 5.9× faster than the leading long‑context baseline.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computation and Language.

arXiv Machine Learning

Jul 9

Final Checkpoints Are Not Enough: Analyzing Latent Reasoning Faithfulness Along Training Trajectories

arXiv:2607. 06648v1 Announce Type: new Abstract: Latent reasoning methods perform multi-step inference entirely in the model's continuous hidden states, promising more compact and efficient reasoning.

By Hengyu Jin, Shu Yang, Di Wang

benchmarks

arXiv AI

Jun 16

Tyler: Typed Latent Reasoning for Language Models -- When to Think, What to Compute, and How Much to Allocate

arXiv:2606. 16360v1 Announce Type: cross Abstract: Chain-of-thought (CoT) prompting improves reasoning in large language models (LLMs) by externalizing intermediate computation as discrete text tokens, but this textual interface also introduces redundancy and inference overhead.

By Hanyu Lin, Min Cai, Jiawei Wen, Haodi Zhang

llms

arXiv AI

Jun 2

Latent Reasoning in TRMs is Secretly a Policy Improvement Operator

arXiv:2511. 16886v5 Announce Type: replace-cross Abstract: Recently, small models with latent recursion have obtained promising results on complex reasoning tasks.

By Arip Asadulaev, Rayan Banerjee, Fakhri Karray, Martin Takac

diffusion reinforcement-learning

arXiv AI

Sep 15

How Many Thoughts Can a Vector Hold? The Capacity of Reasoning by Superposition

The paper investigates how continuous latent states in large language models can store multiple reasoning steps through superposition. It challenges the intuition that retaining only the current reasoning frontier is optimal, showing that cumulative superposition of the full reasoning history can actually require fewer representational dimensions. The authors demonstrate that this approach preserves more valid evidence, improves downstream outcome discrimination, and delays unreliability, while also establishing that uniform cumulative weighting of memories is minimax‑optimal for future reasoning.

By Hongyu Gu, Chang Liu, Jingwen Fu

llms

arXiv Machine Learning

Jun 17

Continual Self-Improvement with Lightweight Experiential Latent Memories

arXiv:2606. 17803v1 Announce Type: new Abstract: Large language models achieve strong reasoning performance by scaling inference-time compute, yet remain fundamentally stateless, discarding the rich, self-produced reasoning traces generated during this process.

By Vaggelis Dorovatas, Nancy Kalaj, Rahaf Aljundi

llms reinforcement-learning benchmarks

arXiv AI

Jul 29

Penelope: Localized Latent Recurrence for Efficient Structured Reasoning

arXiv:2607. 25915v1 Announce Type: new Abstract: Complex structured reasoning tasks often require additional computation, yet current language models obtain it mainly by increasing parameter scale or by serializing intermediate steps as chain-of-thought (CoT) tokens.

By Yutong Chen, Shouqian Shi, Xinran Liu, Haochen Wang, Jiaying Wang, Tianxing Xu, Yuanxi Wang, Zirui Ding

llms diffusion efficiency benchmarks

One email a morning, machine-written

Related stories

Final Checkpoints Are Not Enough: Analyzing Latent Reasoning Faithfulness Along Training Trajectories

Tyler: Typed Latent Reasoning for Language Models -- When to Think, What to Compute, and How Much to Allocate

Latent Reasoning in TRMs is Secretly a Policy Improvement Operator

How Many Thoughts Can a Vector Hold? The Capacity of Reasoning by Superposition

Continual Self-Improvement with Lightweight Experiential Latent Memories

Penelope: Localized Latent Recurrence for Efficient Structured Reasoning