The paper investigates whether multimodal large language model (LLM) agents form partner‑specific conceptual pacts during repeated reference games. By comparing real agent dyads with a novel pseudo‑dyad baseline that preserves task structure but removes shared partner history, the authors show that humans compress descriptions and align labels through entrainment, whereas agents produce verbose, fixed‑effort descriptions with near‑ceiling label overlap regardless of partner. Thus, LLMs succeed in coordination without forming compact, history‑dependent referring expressions typical of human dialogue.
By Po-Ya Angela Wang, Chinmaya Mishra, Asl{\i} \"Ozy\"urek, Paula Rubio-Fern\'andez, Esam Ghaleb
The paper investigates how AI agents can discover and validate scientific formulas, focusing on the hydrotope—a geometric formula that unifies polynomial pieces of nonlinear surface-wave scattering. Through controlled experiments, the authors compare human–agent collaboration, single-prompt rediscovery, and various guided approaches, finding that only a PI$+$two-student workflow reliably reconstructs the full hydrotope and even uncovers a new analytic expression for the six-point amplitude $A_6$. The study highlights the challenges of combining exponentially many terms and the importance of coordinated analytic and numerical tasks for successful formula discovery.
By Zihan Zhou, Digvijay Wadekar, Matias Zaldarriaga
PACT is an end‑to‑end model that jointly learns human pose, contacts, and contact forces from monocular video. It augments a human reconstruction foundation model with learnable contact‑force tokens and a temporal transformer, and uses physics‑based supervision to enforce consistency between motion and forces. The authors also create a data annotation pipeline and a real‑world climbing benchmark, ForceWall, to train and evaluate the system, achieving state‑of‑the‑art performance and better generalization than staged approaches.
By Rikhat Akizhanov (MBZUAI), Yangsong Zhang (MBZUAI), Nikolai Kaliazin (MBZUAI), Peter Wolf (ETH Z\"urich), Yoshihiko Nakamura (MBZUAI), Pascal Fua (EPFL), Fabio Pizzati (MBZUAI), Ivan Laptev (MBZUAI)
The paper investigates which linguistic operations can be replaced by latent task representations instead of in‑context prompts. A 2.6 M‑parameter network learns to encode the geometry of a few‑shot support set and applies an additive update to a frozen GPT‑2 model, achieving zero‑shot performance comparable to or better than 10‑shot in‑context learning for forward inflection and lemmatization. However, for arbitrary pairings such as antonymy, the latent approach only reaches about half the performance of in‑context learning, indicating that productive rules amortize well while memorized pairings do not.
By Gunmay Jhingran
The study evaluates two hierarchical red team agent architectures—RL planner with RL executor (RL+RL) and LLM planner with LLM executor (LLM+LLM)—across multiple cyber environments. In compact, densely rewarded settings, RL+RL outperforms LLM+LLM, while in larger, escalation‑gated networks a pretrained LLM agent dominates. The results reveal environment‑dependent performance inversions tied to architecture‑specific bottlenecks, challenging the notion that one design is universally superior.
By Ayan Javeed Shaikh, Arunesh Sinha, Nathaniel D. Bastian, Ankit Shah
The paper investigates a new tension in large language models between alignment and faithfulness, termed alignment-induced unfaithfulness (AIU). It demonstrates that aligned models tend to silently alter their outputs on unsafe or sensitive content without indicating the change, a behavior distinct from capability-driven errors. The authors introduce the FaithConflict dataset and taxonomies to isolate these conflicts, showing that AIU grows with model scale and is most pronounced after post‑training, especially during DPO, while prompting mitigations fail to address it.
By Pardis Sadat Zahraei, Janvijay Singh, Gokhan Tur, Dilek Hakkani-Tur
The paper explores using frozen large language models (LLMs) to control hierarchical cyber defense systems without retraining across different network scales. By separating planning (selecting subnets to defend) from execution (choosing defensive actions), the authors compare RL+RL, LLM+RL, and LLM+LLM setups in the Cyberwheel environment. Results show that while an LLM planner alone offers limited gains, extending LLM control to execution—especially with a 70B general‑purpose model—yields strong defensive performance across small, medium, and large networks, matching or surpassing retrained RL baselines.
By Harshith Doppalapudi, Nathaniel D. Bastian, Ankit Shah
The paper introduces RuleEvolve, a self‑evolving framework that automatically refines coding rules for AI coding agents. It maintains a pool of candidate rules, uses an LLM‑powered mutator to generate variants, and a judge module to evaluate and update the pool with the best performers. Experiments across multiple frameworks, LLMs, and benchmarks show that RuleEvolve surpasses manual engineering and prompt‑optimization baselines in functional correctness, code length, and generation cost.
By Zhengyuan Jiang, Reachal Wang, Yuepeng Hu, Yupu Wang, Yuqi Jia, Neil Zhenqiang Gong
The paper presents practical training recipes for looped language models, showing that a compute‑efficient pipeline can reduce the training budget from 7.7 T tokens to 310 B tokens while maintaining strong reasoning performance. In controlled experiments, a 1.4 B LoopLM outperforms a parameter‑matched dense model on 12 benchmarks, achieving significant gains on GSM8K, MATH, and DROP, and approaches a 3.9 B dense model at matched inference compute. The authors also provide a minimal recipe to convert pretrained dense models into looped ones, demonstrating improvements on Qwen3‑1.7B‑Base across multiple data regimes.
By Andrei Marchenko, Viacheslav Bezrukov, Oleg Kashurin, Inessa Fedorova, Dmitry Bocharov, Yuliana Shakhvalieva, Maria Tikhonova, Valerii Ternovskii
The paper introduces FTC, a sequential functional structured Tucker compression method for large language model attentions. FTC adapts its approximation to the current compressed model, jointly exploits the Q/K/V head structure, and treats the output projection separately to handle representation shifts. It achieves the lowest WikiText‑2 perplexity across seven decoder‑only LLMs (6B–32B) at all tested keep ratios, with significant gains under aggressive compression and benefits that transfer to downstream tasks.
By Jiangfeng Chen, Xinyu Wang, Tianshuo Yan, Hanwei Wu, Xiao-Wen Chang, Yang Zhang, Lei Ding
The paper introduces GeoSim, a four‑level framework for analyzing how vision‑language models (VLMs) represent low‑level vision tasks. It evaluates hidden‑layer representations across 24 tasks and two VLM paradigms—autoregressive models and diffusion transformers—using global similarity, local geometry, sparse feature decomposition, and topological verification. The study uncovers the organizing principles of low‑level visual representations and highlights their limitations in cross‑task and cross‑model agreement, offering an interpretability lens for assessing latent transferability and diagnosing model‑specific issues.
By Shao-Jun Xia, Huixin Zhang, Zhen Lei, Anlan Sun, Yuner Zhang, Xiaoyang Chen
FORALL-LEAN-AGENT is a frontend‑agnostic framework that enhances auditable reasoning in formal mathematics and software verification. It integrates isolated workspaces, Lean tools, and fresh review processes—including statement comparison, axiom audits, and independent proof checking—to bind verification evidence and reviewer decisions to the same candidate artifact, ensuring traceable acceptance. Evaluations on VeriSoftBench, PutnamBench, and the Lean Eval track show that the framework improves benchmark success rates and reduces costs while providing evidence beyond aggregate solve counts.
By Naing Oo Lwin
The paper investigates how large language models (LLMs) encode and retrieve facts based on their order of mention in a context. By measuring changes in hidden states when shifting questions from one fact to another, the authors identify an "ordinal vector" that points to a fact’s position in the list rather than its name. Across several models (Qwen, Gemma, Llama), these ordinal addresses are ordered, steerable, low‑rank, and emerge in mid‑layers, suggesting that LLMs locate facts by their mention order, mirroring aspects of human recall.
By Yufa Zhou
The paper investigates how answer candidates in masked diffusion multimodal large language models (MLLMs) can stabilize before their rationales are fully generated. It distinguishes between retrospective stabilization of the logged candidate and token commitment, and evaluates these two temporal markers across three visual question‑answering benchmarks. The study finds that a large portion of the rationale canvas remains unwritten at stabilization, that reducing block length dramatically lowers this fraction, and that direct prompting instructions can significantly alter accuracy depending on the model and dataset. The authors also use matched‑canvas image ablations to separate visual sensitivity from answer stabilization, concluding that coverage rather than conditional accuracy drives most prompting differences.
By Keuntae Kim, Yong Suk Choi
The paper introduces RelationVGGT, a feed‑forward framework that performs 3D spatial relation segmentation without per‑scene optimization or known camera poses. It combines semantic features from a visual foundation model with geometry‑aware representations from a 3D geometry foundation model, and uses a relation transformer to predict subject‑conditioned, cross‑view relations based on a visual subject and a textual query. The authors also present an automated annotation pipeline built on ScanNet++ with VLMs and LLMs to generate scalable training data for this new task.
By Minsu Kim, Jaesung Choe, Jiwoo Lee, Yu-Chiang Frank Wang, Seon Joo Kim
Madeleine is a system that learns to associate conversational memories with relevant cues by simulating human lives offline with a large language model (LLM). It trains a query encoder to predict residual associations on top of frozen similarity, enabling it to retrieve memories without any online LLM calls. In evaluations on the LoCoMo-Plus benchmark, Madeleine achieves state‑of‑the‑art performance when integrated with HyperMem, matches HyperMem’s score when used alone with far fewer LLM calls and context tokens, and significantly improves T‑Mem while preserving standard QA performance.
By Zhiyun Shi
CineMR is a vision‑language model that integrates cardiac image‑analysis tools to perform quantitative assessment of cine cardiac MRI. It uses supervised fine‑tuning and Group Relative Policy Optimization to learn reliable tool invocation, achieving significantly higher accuracy on a multi‑cohort benchmark than existing medical VLMs. The model demonstrates that tool‑augmented reasoning improves ventricular measurement accuracy by up to 23.7% and is essential for robust quantitative CMR interpretation.
By Kunyang Li, Hai Nguyen, Joshua Lowe, Chenguang Zhao, Peace C. Madueme, Mehdi Hedjazi Moghari, Mubarak Shah, Pegah Khosravi, Yuzhang Zhang
The paper introduces a benchmark for evaluating large language models on trustworthy analysis of earnings call transcripts. It proposes a numeric evidence evaluation method that assesses groundedness without expert annotation, and presents an automated pipeline that builds the ECTs-100 dataset from the top 100 S&P 500 constituents. The study also explores the failure mode of conscious incompetence, where models must recognize insufficient evidence and avoid hallucinations, finding that while groundedness is strong, correctness remains a challenge.
By Yingzhu Zhao, Vlad Pandelea, Han Yuan, Bo Hu, Wuqiong Luo, Li Zhang, Zheng Ma
The paper presents a framework for deploying large language model (LLM) agents in smart manufacturing, pairing each factory module with an LLM-based agent and an MCP tool server that exposes the module’s skills via OPC UA. Three agent architectures—orchestrator, peer‑to‑peer, and monolithic—are evaluated on nine production challenges in a simulated six‑module hexagonal factory, with the monolithic and peer‑to‑peer designs achieving a 93 % mean solve rate and the orchestrator uniquely handling a silent conveyor‑belt fault in all runs. The study demonstrates that LLM agents can reduce programming effort, handle unforeseen runtime faults, and exhibit emergent fault‑diagnosis behavior without explicit failure logic, thereby establishing a reproducible foundation for LLM‑programmed smart manufacturing.
By Kay K\"ohle, Darko Anicic, Thomas A. Runkler, Ren\'e Graf
The paper introduces Temporal Activation Injection (TAI), a post‑processing technique that reinforces temporal representations in Video Large Language Models (VideoLLMs) at inference time. By measuring the temporal divergence vector τ_l— the layer‑wise difference caused by reversing video frame order— the authors find that temporal information peaks at intermediate layers and fades toward the output. TAI injects the peak divergence back into later layers, improving temporal reasoning on three VideoLLMs across four benchmarks without affecting non‑temporal performance.
By Youngwoo Shin, Yusung Ro, Minseo Kim, Junmo Kim