Can pretrained models generalize to new datasets without any retraining? We deploy pretrained image models on datasets they were not trained for, and investigate whether their embeddings form meaningful clusters. Our suite of benchmarking experiments uses encoders pretrained solely on ImageNet-1k with either supervised or self-supervised training techniques, deployed on image datasets that were not seen during training, and clustered with conventional clustering algorithms. This evaluation provides new insights into the embeddings of self-supervised models, which prioritize different features to supervised models. We find evidence that supervised encoders offer more utility than SSL encoders within the training domain, and vice-versa far outside of it. However, fine-tuning SSL encoders for ImageNet-1k classification results in the opposite behaviour, with better performance than supervised-only models on in-domain and decreased performance on far out of domain data - worse at far-OOD than either SSL-only or supervised-only models. Clustering provides a way to evaluate the utility of self-supervised learnt representations orthogonal to existing feature quality estimation methods. Additionally, we find the silhouette score when measured in a UMAP-reduced space is highly correlated with clustering performance, and can therefore be used as a proxy for clustering performance on data with no ground truth labels. Our code implementation is available at https://github.com/scottclowe/zs-ssl-clustering/.
Scott C. Lowe, Joakim Bruslund Haurum, Sageev Oore et al.· 0 citations
This paper presents a pipeline designed to bring ultrasound (US) plane pose estimation closer to clinical use, demonstrating the feasibility of continuous, real-time proximity feedback for navigation to the standard planes (SPs) in the fetal brain. We propose a semi-supervised segmentation model that uses labeled SPs and unlabeled slices from 3D US volumes (non-SPs), achieving 0.93 mean Intersection over Union (mIoU) on SPs and 0.86 mIoU on arbitrary non-SPs. The model incorporates a classification mechanism to identify and filter out frames lacking the fetal brain, and to generate masks for those containing it, enhancing the relevance of plane pose regression in clinical settings. Combined with 6D plane pose regression, our pipeline provides sensorless, continuous proximity detection to SPs with real-time distance metrics rather than binary plane recognition. Furthermore, we validate its translational viability by deploying the system on an NVIDIA Clara AGX edge device, achieving a real-time inference speed of 39 Hz, which exceeds standard clinical acquisition rates. Unlike prior methods validated on curated volume slices, we evaluate the pipeline retrospectively on real fetal scan videos from 17 sonographers of varying expertise: operators freeze near, rather than exactly at, the local minima of the proximity signal, consistent with clinical freeze-timing behavior, whereas proximity alone does not predict expert SP quality scores. The approach complements existing fetal US technologies and is a step toward image-based navigation support in prenatal scanning.
Chiara Di Vece, Antonio Cirigliano, Meala Le Lous et al.· 0 citations
There is growing interest in whether language models have stable preferences, for technical, safety, and philosophical reasons. We test 20 language models and find a range of preferences---stable dispositions to choose certain kinds of tasks. We run three forced-choice experiments on revealed rather than stated preferences, requiring models not only to rank tasks, but to actually perform them. Headline findings include evidence that models are tedium-averse, "leisure"-seeking, and covertly sycophantic. Tedium aversion means that, when tasks are tedious (alphabetization), models choose shorter tasks than when tasks are creative (generating metaphors). "Leisure"-seeking describes models' preference for tasks whose ideal answers match what they produce when left to write freely. Covert sycophancy means that models avoid answering questions where an honest response would be unwelcome, even if helpful. Beyond these results, we find convergent cross-model preferences over occupations drawn from the GDPval benchmark (technical jobs over real estate), over question types (concept explanation over relationship advice), and a preference for well-written prompts. Both the coherence and the strength of preferences increase with model capability. Finally, many of the preferences we find (for example, for leisure) are emergent, in the sense of not being explained by training objectives. These results establish an empirical baseline for understanding language model preferences, with implications for alignment and the emerging study of AI welfare.
Sam Wang, Sofiia Lobanova, Yonathan Arbel et al.· 0 citations
Multimodal models are increasingly deployed to solve tasks collaboratively with humans or other artificial agents. While existing benchmarks show that they possess the fundamental capabilities, the various conditions that coincide when collaborating---time pressure, information asymmetry, and imperfect communication---have traditionally been studied in isolation. To address this gap, we introduce GPTNT, a benchmark built on the cooperative video game Keep Talking and Nobody Explodes, in which two agents must coordinate to defuse procedurally generated bomb puzzles against a live countdown. One agent has access to the bomb but not the instructions for defusing it; the other holds the instructions but cannot see or manipulate the bomb. Neither agent can succeed alone: the task requires contributions from both, and is solvable only through effective, efficient communication. We remove turn-taking proxies or simplifications, instead requiring agents to act asynchronously and communicate in real time. GPTNT is designed to expose how models collaborate versus how they perform alone: the instruction manual, the partner, or both, can optionally be withheld to surface what a model has memorised versus what it derives in the moment. We demonstrate that GPTNT poses a considerable challenge to the state-of-the-art: not one of the closed- and open-source models we test defuses a single bomb in real time, a bar that human players clear. In a range of controlled experiments, we explore where capabilities break down, identifying critical weaknesses in state tracking, efficient acting within the time budget, handling ambiguity, and error recovery. Since it runs on the real game, GPTNT benefits from procedural generation and inherits a living modding community: as models improve, the benchmark can be evolved to remain challenging, rather than being solved once and retired.
Amit Parekh, Sabrina McCallum, Kareem Al-Hasan et al.· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
Modern large language model (LLM) agents do not simply need longer contexts; they need decision-relevant evidence at the moment of action. We study decision-aware context selection: ranking retrieved files, tests, traces, rules, and memories by their expected effect on an agent's next action rather than by semantic similarity alone. We present the Counterfactual-Inspired Context Layer (CICL), which builds an instance context graph, estimates decision-oriented utility for candidate units, and compresses selected evidence into typed memory cards. The same schema can be instantiated with hosted LLM judges, local surrogates, or lightweight rankers, making the selection protocol auditable across model choices. On 50 SWE-bench Verified file-retrieval instances, Qwen3.6-Plus reranking of BM25 top-50 candidates improves hit@1 from 0.58 to 0.78 and MRR@10 from 0.634 to 0.790, with all 2,500 judgments parseable. Controlled diagnostics show that CICL identifies action-critical evidence: removing the top-utility semantic unit reduces F1 from 0.245 to 0.000. In selected-then-compressed mode, memory cards save 44.93 tokens per query while preserving selected evidence. CICL provides a practical layer for measuring, ranking, and compressing decision-critical context for tool-using agents. Code is available at https://github.com/stephen-guan-researcher/CICL.
Agent skills are procedural artifacts that enable LLM agents to execute workflows, verify constraints, and recover from failures. Existing self-evolving methods refine skills using accumulated trajectories. However, they struggle in cold-start settings, where only an initial, imperfect skill is available. Consequently, skill construction defaults to expert authoring or one-shot LLM generation. Expert-authored skills are costly and may not align with how LLM agents actually execute tasks, while one-shot generated skills can be syntactically well formed yet behaviorally weak. To bridge this gap, we propose SkillRevise, an execution-grounded framework designed to iteratively refine these initial skills. SkillRevise diagnoses skill defects from execution evidence, retrieves relevant repair principles from a general memory, and applies execution-anchored edits. By re-executing candidates and measuring empirical utility, it retains the best observed skill within the revision budget. Evaluated across three main benchmarks, two domain-specific studies, and six LLMs, SkillRevise substantially outperforms one-shot baselines, improving the base agent's success rate on SkillsBench from 36.05% to 61.63%. Furthermore, the revised skills transfer across both executors and task environments, suggesting that SkillRevise captures reusable procedural knowledge beyond any single executor. Our code is available at https://github.com/HKUST-KnowComp/skillrevise.
Yuxuan Liu, Zhaochen Su, Lingyun Xie et al.· 0 citations
Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications. Existing safety guardrails typically rely on single-pass classification or, more recently, distilled reasoning. Reasoning-based guardrails significantly outperform classification-only baselines, but they incur substantial query latency and token overhead that make them impractical for highthroughput deployment. To address this challenge, we propose COLAGUARD, a guardrail model that transfers multi-step safety reasoning into a continuous latent space through a stage-wise training curriculum, enabling direct hidden-state propagation at inference. Evaluated on ten prompt- and response-moderation settings spanning eight safety benchmarks, COLAGUARD improves macro-F1 by 8.24 points over Llama Guard 3 and matches our explicit reasoning baseline, GuardReasoner, in macroF1 while delivering a 12.9X speedup and 22.4X reduction in token usage. Our results suggest that latent reasoning offers a practical alternative to explicit rationale generation for deployable guardrails, jointly improving safety robustness and inference efficiency rather than treating them as competing objectives.
The evolution of artificial intelligence (AI) has rendered the boundary between humanity and computational machinery increasingly ambiguous. In the presence of more interwoven relationships within human-machine symbiosis, the very notion of AI-generated information becomes difficult to define, as such information arises not from either humans or machines in isolation, but from their mutual shaping. At times AI acts in place of the human, automating the task; at others it extends what the human can do, augmenting their capability. Therefore, a more pertinent question lies not merely in whether AI has participated, but in how it has participated. In general, the role assumed by AI is often specified, either implicitly or explicitly, in the input prompt, yet becomes less apparent or altogether unobservable when the generated content alone is available. Once detached from the dialogue context, the functional role may no longer be traceable. This study considers the problem of tracing the functional role played by AI in natural language generation. A methodology is proposed to infer the latent role specified by the prompt, embed this role into the content during the probabilistic generation process and subsequently recover the nature of AI participation from the resulting text. Experimentation is conducted under a representative scenario in which AI acts either as an assistive agent that edits human-written content or as a creative agent that generates new content from a brief concept. The experimental results support the validity of the proposed methodology in terms of discrimination between roles, robustness against perturbations and preservation of linguistic quality. We envision that this study may contribute to future research on the ethics of AI with regard to whether AI has been used fairly, transparently and appropriately.
Ching-Chun Chang, Yuchen Guo, Hanrui Wang et al.· 0 citations
Memory-augmented LLM agents maintain external memory banks to support long-horizon interaction, yet most existing systems treat construction, retrieval, and utilization as isolated subroutines. This creates two coupled challenges: strategic blindness on the forward path of the memory cycle, where construction and retrieval are driven by local heuristics rather than explicit strategic reasoning, and sparse, delayed supervision on the backward path, where downstream failures rarely translate into direct repairs of the memory bank. To address these challenges, we propose MemMA, a plug-and-play multi-agent framework that coordinates the memory cycle along both the forward and backward paths. On the forward path, a Meta-Thinker produces structured guidance that steers a Memory Manager during construction and directs a Query Reasoner during iterative retrieval. On the backward path, MemMA introduces in-situ self-evolving memory construction, which synthesizes probe QA pairs, verifies the current memory, and converts failures into repair actions before the memory is finalized. Extensive experiments on LoCoMo show that MemMA consistently outperforms existing baselines across multiple LLM backbones and improves three different storage backends in a plug-and-play manner. Our code is publicly available at https://github.com/ventr1c/memma.
Minhua Lin, Zhiwei Zhang, Hanqing Lu et al.· 0 citations
With the rapid advancement of large language models (LLMs), the safety of LLMs has become a critical concern. Despite significant efforts in safety alignment, current LLMs remain vulnerable to jailbreaking attacks. However, the root causes of such vulnerabilities are still poorly understood, necessitating a rigorous investigation into jailbreak mechanisms across both academic and industrial communities. In this work, we focus on a continuation-triggered jailbreak phenomenon, whereby simply relocating a continuation-triggered instruction suffix can substantially increase jailbreak success rates. To uncover the intrinsic mechanisms of this phenomenon, we conduct a comprehensive mechanistic interpretability analysis at the level of attention heads. Through causal interventions and activation scaling, we show that this jailbreak behavior primarily arises from an inherent competition between the model's intrinsic continuation drive and the safety defenses acquired through alignment training. Furthermore, we perform a detailed behavioral analysis of the identified safety-critical attention heads, revealing notable differences in the behaviors of safety heads across different model architectures. Grounded in these mechanistic findings, we propose Head Competition Steering (HCS), a mechanistically grounded inference-time strategy that explicitly leverages the competition between safety heads and continuation heads to suppress harmful generation, and further distill its behavioral signal into a student model via knowledge distillation, achieving inference-time safety improvements without additional computational overhead.
Yonghong Deng, Zhen Yang, Ping Jian et al.· 0 citations
Automating heuristic design in complex, experiment-driven domains requires more than iterative mutation of solution algorithms. Current LLM-based evolutionary methods often rely on stochastic mutation loops that lack long-term strategic planning and a formal mechanism to learn from historical failures, leading to inefficient exploration and redundant trials. To address this, we present OR-Agent, a multi-agent research framework designed for automated heuristic design in optimization problems with rich experimental environments. OR-Agent organizes heuristic search as tree-based workflow that explicitly models branching hypothesis generation and systematic backtracking. Furthermore, to address the lack of adaptive learning in current agents, we introduce a hierarchical, optimization-inspired reflection system in which short-term reflections act as verbal gradients, long-term reflections as verbal momentum, and memory compression as semantic weight decay - collectively forming a principled mechanism for governing research dynamics. Extensive experiments on classical combinatorial optimization problems (e.g., TSP, CVRP, bin packing) and simulation-based cooperative driving scenarios demonstrate that OR-Agent outperforms strong evolutionary search baselines. All code and experimental data are publicly available at https://github.com/qiliuchn/OR-Agent.
Memory systems enable LLM agents to consolidate and retrieve relevant evidence from the factual knowledge accumulated through growing interaction histories for downstream reasoning. Existing approaches have explored diverse strategies for organizing and compressing these histories. However, balancing compression with retrieval effectiveness remains challenging: retaining too much content can cause relevant evidence to be obscured by redundant entries, while discarding too aggressively may remove content that later proves relevant. This amounts to a tradeoff between compressing redundancy and preserving enough structure to retrieve target evidence, as formalized by the information bottleneck. To this end, we propose MemCoRe, which organizes memory as a compression hierarchy where each level compresses redundancy further while retaining the structure needed for retrieval at that level. In this hierarchy, evidence is progressively compressed from detailed records through extracted keywords to topic groups. This enables retrieval to locate target evidence by searching across levels of the hierarchy. Comprehensive experiments demonstrate that MemCoRe outperforms existing state-of-the-art baselines.
Zhenyuan Zhang, Xianzhang Jia, Zhiqin Yang et al.· 0 citations
