Category

Agent-Editing World Model: Rethinking World Modeling for LLM Agents

Edits an LLM agent's internal task state and action continuations to remove unsupported assumptions and noisy plans, improving downstream decisions instead of reconstructing high-entropy tool outputs. Combines Action Judge, State Revision, and EditAct and shows consistent gains across benchmarks and agent backbones.

Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs

Shows that Transformer LLMs output a superposition of next-token distributions when token embeddings from distinct text streams are linearly combined, finds this property is architectural but weakened by pretraining, and demonstrates it can be restored by lightweight fine-tuning and disentangled at decode time.

Post-Training Leaves Behavioral Shadows on Unrelated Decisions

Shows that private post-training changes leave measurable "behavioral shadows" in task-unrelated single-word outputs and proposes Active Taskless Distillation (ATD) to transfer capabilities to a student using only those single-word teacher responses, without teacher logits or parameters.

Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge

Teaches small reasoning models to diagnose when additional internal thinking is insufficient and to selectively query stronger models; introduces FlyBy, a supervised + cost-aware RL framework that learns when/what to ask. Improves pass@ metrics on hard benchmarks while reducing serving cost.

Improving Test-Time Scaling with Adaptive Looped Transformers

Improves test-time scaling of looped transformers by adaptively assigning extra recurrent iterations to tokens that benefit most. TaH2 is a post-training method that jointly trains an iteration decider with the backbone using lookahead depth supervision, boosting the accuracy–compute slope and peak accuracy on challenging benchmarks.

E-MoE: Enhanced Mixture-of-Experts for Non-Factorized Diffusion Language Models

Models sequence generation by unmasking multiple tokens per denoising step and replaces a factorized reverse process with a mixture over discrete routing-based latents from an MoE backbone; improves few-step sampling quality without increasing active parameters.

A Missing Piece for Trustworthy AI Reviewers: From Benchmarking Rhetorical Robustness to SciCore Review

Assesses how AI reviewers react to content-preserving rewrites and introduces RobustReview, a controlled benchmark plus SciCore, a dual-branch reviewer that combines full-manuscript judgment with a structured 'science core' summary to reduce rhetorical sensitivity.

CANOPY: Adaptive-Granularity Evidence Compression for Multimodal RAG

Adaptive-granularity post-retrieval evidence compressor for multimodal RAG that selects variable-sized regions from retrieved text, tables, images, and videos to retain necessary context while cutting irrelevant content. Key features include a hierarchy-based node encoder with parent-relative refinement and a critic that issues targeted follow-up retrievals; yields higher QA accuracy across five benchmarks and reduces reader-input tokens by ~14–28%.

Hierarchical Continuous Diffusion Language Models

Couples a continuous latent diffusion trajectory with discrete token readouts so tokens are read from the latent at every step and fed back as a scaffold, reducing token-independence in parallel decoding and improving structured-task accuracy and LM perplexity.

Source Preference in the Wild: How LLM Agents Favor Items by Source, and How to Reduce It

Analyzes how LLM agents prefer items from particular sources during end-to-end search and how these preferences affect selections across shopping, accommodation, and scholarly domains. Shows source labels can override item quality and evaluates mitigation strategies such as hiding sources, relabeling, supplying missing information, and counter-prompts.

ALoDLM: Adaptively Looped Diffusion Language Models

Introduces a token-adaptive latent recurrence for diffusion language models that allocates computation per token during denoising. Key features: iterative latent refinement, discrete-feedback commits, and learned token-wise schedules to improve parallel decoding quality-efficiency trade-offs.

Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability

Analyzes cross-tokenizer on-policy distillation (OPD) and shows that focusing supervision on strictly aligned token positions—or a student-selected top-16 subset of the shared vocabulary—retains most distillation gains, while adding span-level MSE supervision reduces downstream accuracy.