Recent proposals for regulating frontier AI models have sparked concerns about the cost of safety regulation, and most such regulations have been shelved due to the safety-innovation tradeoff. This paper argues for an alternative regulatory approach that ensures AI safety while actively promoting innovation: mandating that large AI laboratories release small, openly accessible analog models (scaled-down versions) trained similarly to and distilled from their largest proprietary models. Analog models serve as public proxies, allowing broad participation in safety verification, interpretability research, and algorithmic transparency without forcing labs to disclose their full-scale models. Recent research demonstrates that safety and interpretability methods developed using these smaller models generalize effectively to frontier-scale systems. By enabling the wider research community to directly investigate and innovate upon accessible analogs, our policy substantially reduces the regulatory burden and accelerates safety advancements. This mandate promises minimal additional costs, leveraging reusable resources like data and infrastructure, while significantly contributing to the public good. Our hope is not only that this policy be adopted, but that it illustrates a broader principle supporting fundamental research in machine learning: deeper understanding of models relaxes the safety-innovation tradeoff and lets us have more of both.

Fixing bugs in large programs is a challenging task that demands substantial time and effort. Once a bug is found, it is reported to the project maintainers, who work with the reporter to fix it and eventually close the issue. However, across the program, there are often similar code segments, which may also contain the bug, but were missed during discovery. Finding and fixing each recurring bug instance individually is labor intensive. Even more concerning, bug reports can inadvertently widen the attack surface as they provide attackers with an exploitable pattern that may be unresolved in other parts of the program. In this paper, we explore these Recurring Pattern Bugs (RPBs) that appear repeatedly across various code segments of a program or even in different programs, stemming from a same root cause, but are unresolved. Our investigation reveals that RPBs are widespread and can significantly compromise the security of software programs. This paper introduces BugStone, a program analysis system empowered by LLVM and a Large Language Model (LLM). The key observation is that many RPBs have one patched instance, which can be leveraged to identify a consistent error pattern, such as a specific API misuse. By examining the entire program for this pattern, it is possible to identify similar sections of code that may be vulnerable. Starting with 135 unique RPBs, BugStone identified more than 22K new potential issues in the Linux kernel. Manual analysis of 400 of these findings confirmed that 246 were valid. We also created a dataset from over 1.9K security bugs reported by 23 recent top-tier conference works. We manually annotate the dataset, identify 80 recurring patterns and 850 corresponding fixes. Even with a cost-efficient model choice, BugStone achieved 92.2% precision and 79.1% pairwise accuracy on the dataset.

Large language models (LLMs) can exhibit biases in reasoning capabilities due to linguistic modality, performing better on tasks in one language versus another, even with similar content. Most previous works evaluate this through reasoning tasks where reliance on strategies or knowledge can ensure success, such as in commonsense or math tasks. However, abstract reasoning is vital to reasoning for everyday life, where people apply "out-of-the-box thinking" to identify and use patterns for solutions, without a reliance on formulaic approaches. Comparatively, little work has evaluated linguistic biases in this task type. In this paper, we propose a task inspired by the New York Times Connections: GlobalGroup, that evaluates models in an abstract reasoning task across several languages. We constructed a game benchmark with five linguistic backgrounds -- English, Spanish, Chinese, Hindi, and Arabic -- in both the native language and an English translation for comparison. We also proposed game difficulty measurements to evaluate models on games with similar difficulty, enabling a more controlled comparison, which is particularly important in reasoning evaluations. Through experimentation, we find English modalities largely lead to better performance in this abstract reasoning task, and performance disparities between open- and closed-source models.

Transformers, powered by the attention mechanism, are the backbone of most foundation models, yet they suffer from quadratic complexity and difficulties in dealing with long-range dependencies in the input sequence. Recent work has shown that state space models (SSMs) provide an efficient alternative, with the S6 module at the core of the Mamba architecture achieving state-of-the-art results on long-sequence benchmarks. In this paper, we introduce the COFFEE (COntext From FEEdback) model, a novel time-varying SSM that incorporates state feedback to enable context-dependent selectivity, while still allowing for parallel implementation. Whereas the selectivity mechanism of S6 only depends on the current input, COFFEE computes it from the internal state, which serves as a compact representation of the sequence history. This shift allows the model to regulate its dynamics based on accumulated context, improving its ability to capture long-range dependencies. In addition to state feedback, we employ an efficient model parametrization that removes redundancies present in S6 and leads to a more compact and trainable formulation. On the induction head task, COFFEE achieves near-perfect accuracy with two orders of magnitude fewer parameters and training sequences compared to S6. On MNIST, COFFEE largely outperforms S6 within the same architecture, reaching 97% accuracy with only 3585 parameters. These results showcase the role of state feedback as a key mechanism for building scalable and efficient sequence models.

Sparsely-activated Mixture-of-Experts (SMoE) models offer efficient pre-training and low latency but their large parameter counts create significant memory overhead, motivating research into expert compression. Contrary to recent findings favouring expert merging on discriminative benchmarks, we demonstrate that expert pruning is a superior strategy for generative tasks. We prove that merging introduces an irreducible error by causing a "functional subspace collapse", due to the loss of the router's independent, input-dependent control over experts. Leveraging this insight, we propose Router-weighted Expert Activation Pruning (REAP), a novel pruning criterion that considers both router gate-values and expert activation norms. Across a diverse set of SMoE models ranging from 20B to 1T parameters, REAP consistently outperforms merging and other pruning methods on generative benchmarks, especially at 50% compression. Notably, our method achieves near-lossless compression on code generation and tool-calling tasks with Qwen3-Coder-480B and Kimi-K2, even after pruning 50% of experts.

Remote sensing has become a vital tool across sectors such as urban planning, environmental monitoring, and disaster response. While the volume of data generated has increased significantly, traditional vision models are often constrained by the requirement for extensive domain-specific labelled data and their limited ability to understand the context within complex environments. Vision Language Models offer a complementary approach by integrating visual and textual data; however, their application to remote sensing remains underexplored, particularly given their generalist nature. This work investigates the combination of vision models and VLMs to enhance image analysis in remote sensing, with a focus on aircraft detection and scene understanding. The integration of YOLO with VLMs such as LLaVA, ChatGPT, and Gemini aims to achieve more accurate and contextually aware image interpretation. Performance is evaluated on both labelled and unlabelled remote sensing data, as well as degraded image scenarios which are crucial for remote sensing. The findings show an average MAE improvement of 48.46% across models in the accuracy of aircraft detection and counting, especially in challenging conditions, in both raw and degraded scenarios. A 6.17% improvement in CLIPScore for comprehensive understanding of remote sensing images is obtained. The proposed approach combining traditional vision models and VLMs paves the way for more advanced and efficient remote sensing image analysis, especially in few-shot learning scenarios.

As Large Language Models (LLMs) become increasingly integrated into software development workflows, they also become prime targets for adversarial attacks. Among these, backdoor attacks are a significant threat, allowing attackers to manipulate model outputs through hidden triggers embedded in training data. Detecting such backdoors remains a challenge, and one promising approach is the use of Spectral Signature defense methods that identify poisoned data by analyzing feature representations through eigenvectors. While some prior works have explored Spectral Signatures for backdoor detection in neural networks, recent studies suggest that these methods may not be optimally effective for code models. In this paper, we revisit the applicability of Spectral Signature-based defenses in the context of backdoor attacks on code models. We systematically evaluate their effectiveness under various attack scenarios and defense configurations, analyzing their strengths and limitations. We found that the widely used setting of Spectral Signature in code backdoor detection is often suboptimal. Hence, we explored the impact of different settings of the key factors. We discovered a new proxy metric that can more accurately estimate the actual performance of Spectral Signature without model retraining after the defense.

Causal learning is the cognitive process of developing the capability of making causal inferences based on available information, often guided by normative principles. This process is prone to errors and biases, such as the illusion of causality, in which people perceive a causal relationship between two variables despite lacking supporting evidence. This cognitive bias has been proposed to underlie many societal problems, including social prejudice, stereotype formation, misinformation, and superstitious thinking. In this work, we examine whether large language models are prone to developing causal illusions when faced with a classic cognitive science paradigm: the contingency judgment task. To investigate this, we constructed a dataset of 1,000 null contingency scenarios (in which the available information is not sufficient to establish a causal relationship between variables) within medical contexts and prompted LLMs to evaluate the effectiveness of potential causes. Our findings show that all evaluated models systematically inferred unwarranted causal relationships, revealing a strong susceptibility to the illusion of causality. While there is ongoing debate about whether LLMs genuinely understand causality or merely reproduce causal language without true comprehension, our findings support the latter hypothesis and raise concerns about the use of language models in domains where accurate causal reasoning is essential for informed decision-making.

State-of-the-art (SOTA) Automatic Speech Recognition (ASR) systems primarily rely on acoustic information while disregarding additional multi-modal context. However, visual information are essential in disambiguation and adaptation. While most work focus on speaker images to handle noise conditions, this work also focuses on integrating presentation slides for the use cases of scientific presentation. In a first step, we create a benchmark for multi-modal presentation including an automatic analysis of transcribing domain-specific terminology. Next, we explore methods for augmenting speech models with multi-modal information. We mitigate the lack of datasets with accompanying slides by a suitable approach of data augmentation. Finally, we train a model using the augmented dataset, resulting in a relative reduction in word error rate of approximately 34%, across all words and 35%, for domain-specific terms compared to the baseline model.

Large language models (LLMs) are increasingly applied in biomedical domains, yet their reliability in drug-safety prediction remains underexplored. In this work, we investigate whether LLMs incorporate socio-demographic information into adverse event (AE) predictions, despite such attributes being clinically irrelevant. Using structured data from the United States Food and Drug Administration Adverse Event Reporting System (FAERS) and a persona-based evaluation framework, we assess two state-of-the-art models, ChatGPT-4o and Bio-Medical-Llama-3.8B, across diverse personas defined by education, marital status, employment, insurance, language, housing stability, and religion. We further evaluate performance across three user roles (general practitioner, specialist, patient) to reflect real-world deployment scenarios where commercial systems often differentiate access by user type. Our results reveal systematic disparities in AE prediction accuracy. Disadvantaged groups (e.g., low education, unstable housing) were frequently assigned higher predicted AE likelihoods than more privileged groups (e.g., postgraduate-educated, privately insured). Beyond outcome disparities, we identify two distinct modes of bias: explicit bias, where incorrect predictions directly reference persona attributes in reasoning traces, and implicit bias, where predictions are inconsistent, yet personas are not explicitly mentioned. These findings expose critical risks in applying LLMs to pharmacovigilance and highlight the urgent need for fairness-aware evaluation protocols and mitigation strategies before clinical deployment.