Model releases, architecture work and prompting research on large language models — from frontier-lab announcements to the arXiv papers behind them.
arXiv:2602.21223v2 Announce Type: replace Abstract: It is not only what we ask large language models (LLMs) to do that matters, but also how we ask them. Phrases like ``This is urgent'' or ``As your...
arXiv:2603.12050v3 Announce Type: replace Abstract: Translated texts exhibit systematic differences from comparable texts originally written in the target language. Explaining this phenomenon, common...
arXiv:2606.12234v2 Announce Type: replace Abstract: Controlling the output of Large Language Models (LLMs) is a central challenge for their reliable deployment, yet a clear understanding of the invol...
arXiv:2609.12408v1 Announce Type: new Abstract: Hazardous-freight operations at petrochemical logistics nodes are safety-critical for intelligent transportation systems, yet existing vision benchmark...
arXiv:2609.12678v1 Announce Type: new Abstract: Short-video platforms have become a primary news source for the public, which has also enabled the widespread dissemination of fake news videos. We stu...
arXiv:2609.12818v1 Announce Type: new Abstract: Long video understanding often behaves like a visual needle-in-a-haystack problem: query-relevant evidence is sparsely distributed across long temporal...
arXiv:2609.13012v1 Announce Type: new Abstract: Vision-language models retain a substantial amount of pixel-decodable visual content in their visual key-value cache. We show, in our setting, that thi...
arXiv:2504.04903v3 Announce Type: replace Abstract: We present Lunima-OmniLV (abbreviated as OmniLV), a universal multimodal multi-task framework for low-level vision that addresses over 100 sub-task...
arXiv:2606.10468v2 Announce Type: replace Abstract: Coastline detection in remotely sensed imagery is commonly formulated as pixel-wise segmentation, even though coastlines used in coastal monitoring...
arXiv:2607.01630v2 Announce Type: replace Abstract: Dynamic expansion methods for class-incremental learning (CIL) protect task-specific knowledge by growing dedicated tokens or subnetworks, yet our...
arXiv:2609. 12994v1 Announce Type: new Abstract: Heavy-tailed empirical spectral densities of neural-network weight matrices are widely used as diagnostics of implicit self-regularization, but the step complexity of heavy-tail emergence remains poorly understood.
The paper introduces PACEvolve, a framework that improves self‑evolving agents powered by Large Language Models by addressing their tendency to become trapped in local contexts and repeat flawed hypotheses. It does so through three techniques: Hierarchical Context Management to prune memory, Momentum‑Based Backtracking to escape local minima, and a self‑adaptive Collaborative Evolution policy to balance refinement and knowledge transfer. These methods enable the agents to maintain a global view of search momentum and achieve state‑of‑the‑art results on complex evolutionary benchmarks.
The paper introduces a style-aware paraphrasing method for text anonymization that leverages pretrained large language models to build compact stylistic profiles from minimal samples and rewrite text to suppress identifiable style markers while preserving meaning. It demonstrates that this approach reduces authorship attribution F1 scores by 60‑70% on blog and review datasets, outperforming both differential privacy‑based and non‑DP baselines, and maintains content quality and readability.
The paper investigates why Transformers struggle more with structural than lexical compositional generalisation. It argues that this disparity stems from low structural type diversity rather than an inherent limitation of Transformers. By creating linguistically diverse variants of the COGS and SLOG datasets, the authors show that type diversity correlates equally with generalisation in both lexical and structural cases, challenging previous explanations of the difficulty.
The paper introduces Process-aware Language Cognitive Diagnosis (PLCD), a framework that replaces traditional ID-based embeddings in Cognitive Diagnosis Models with language-derived structures and response records. PLCD employs large language models to build concept schemas and cognitive process graphs, and uses a Language-to-Cognition Mapper with DA-MoE experts and contrastive learning to map textual evidence into a unified cognitive space. Experiments demonstrate that PLCD outperforms conventional baselines in student performance prediction and shows strong cognitive transfer, improving cold-start robustness and cognitive grounding.
The paper introduces a belief‑shift evaluation benchmark for large language models (LLMs) using the social‑deduction game Werewolf. By annotating suspicion and accusation messages in LLM‑played games, the authors measure how a village‑side model’s beliefs change after each message, evaluating 40 open‑weight LLMs on 1,224 annotated messages. Results show that larger models better distinguish wolves from villagers, yet accusations still heavily sway beliefs, especially when the accuser is trusted, and even when the accuser is wolf‑aligned. "whyItMatters":"The study highlights that current open‑weight LLMs up to 120B parameters still struggle to integrate accusation content with source trust in strategic communication, revealing limitations in their belief‑updating capabilities in complex social contexts."
The paper introduces Confidence-Gated Transductive Test Generation (CoTT), a method that combines an efficient inductive approach with selective transductive generation triggered only when inductive confidence is low. This adaptive strategy enhances the reliability of test case outputs while limiting additional computational effort. Experiments on code reranking benchmarks demonstrate that CoTT surpasses previous baselines in performance metrics and reduces overall cost compared to applying transductive generation to every input.
Semantically Aligned Gradient-Driven Context-Preserving Image Editing (IABEdit) is a model‑agnostic framework that embeds differentiable semantic verification into the training of generative image editors. By using a frozen vision‑language model to extract spatially‑aware descriptors from ground‑truth edits and a trainable aligner to reproduce them from generated outputs, the residual becomes a gradient that teaches the generator both what to edit and where, without adding inference‑time VLM cost. IABEdit is compatible with various backbones (e.g., U‑Net in Stable Diffusion and MMDiT in FLUX) and improves structural fidelity on MagicBrush, achieves state‑of‑the‑art instruction adherence on RealEdit and EMU Edit, and outperforms the proprietary Gemini agent on the D‑LORD surveillance benchmark under heavy occlusion. "whyItMatters":"IABEdit demonstrates that incorporating semantic verification during training can produce more accurate, well‑localized edits and outperform existing methods even in challenging surveillance scenarios, as shown by its superior metrics and human/GPT‑4o evaluations."
BodhiPromptShield is a policy‑aware mediation layer for LLM agent pipelines that detects sensitive text spans before they propagate, replacing them with typed placeholders, semantic abstractions, or secure tokens and restoring them only at authorized execution boundaries. In evaluations on AI4Privacy, PrivacyLens, and AgentDojo datasets, the system reduces identifier exposure to 7.4% and 1.8% respectively, and limits exact identifier leakage in final actions to 2.1–3.1%. While mediation preserves factual content according to automated metrics, human annotations show a significant drop in inferability from 100% to 24–53%, indicating the need for human validation of semantic‑leakage measures.
The paper investigates how fully autonomous machine‑learning research systems can tackle open‑ended, industry‑grade problems, using telecom ticket retrieval as a case study. It finds that while autonomous research excels at hyperparameter tuning, it lacks human intuition and creativity, yet can achieve about 90% of state‑of‑the‑art performance in a fraction of the time and at modest cost. The authors recommend a hybrid approach where human researchers collaborate with autonomous frameworks for optimal results.