As AI systems become more capable, integrated, and widespread, understanding the associated risks becomes increasingly important. This paper maps the full spectrum of AI risks, from current harms affecting individual users to existential threats that could endanger humanity's survival. We organize these risks into three main causal categories. Misuse risks, which occur when people deliberately use AI for harmful purposes - creating bioweapons, launching cyberattacks, adversarial AI attacks or deploying lethal autonomous weapons. Misalignment risks happen when AI systems pursue outcomes that conflict with human values, irrespective of developer intentions. This includes risks arising through specification gaming (reward hacking), scheming and power-seeking tendencies in pursuit of long-term strategic goals. Systemic risks, which arise when AI integrates into complex social systems in ways that gradually undermine human agency - concentrating power, accelerating political and economic disempowerment, creating overdependence that leads to human enfeeblement, or irreversibly locking in current values curtailing future moral progress. Beyond these core categories, we identify risk amplifiers - competitive pressures, accidents, corporate indifference, and coordination failures - that make all risks more likely and severe. Throughout, we connect today's existing risks and empirically observable AI behaviors to plausible future outcomes, demonstrating how existing trends could escalate to catastrophic outcomes. Our goal is to help readers understand the complete landscape of AI risks. Good futures are possible, but they don't happen by default. Navigating these challenges will require unprecedented coordination, but an extraordinary future awaits if we do.

Large Language Models (LLMs) have shown promising results across various tasks, yet their reasoning capabilities remain a fundamental challenge. Developing AI systems with strong reasoning capabilities is regarded as a crucial milestone in the pursuit of Artificial General Intelligence (AGI) and has garnered considerable attention from both academia and industry. V ar-ious techniques have been explored to enhance the reasoning capabilities of LLMs, with neuro-symbolic approaches being a particularly promising way. This paper comprehensively reviews recent developments in neuro-symbolic approaches for enhancing LLM reasoning. We first present a formalization of reasoning tasks and give a brief introduction to the neuro-symbolic learning paradigm. Then, we discuss neuro-symbolic methods for improving the reasoning capabilities of LLMs from three perspectives: Symbolic LLM, LLM Symbolic, and LLM+ Symbolic . Finally, we discuss several key challenges and promising future directions. We have also released a GitHub repository including papers and resources related to this survey: https://github.com/LAMDASZ-ML/A

Combined Algorithm Selection and Hyperparameter Optimization (CASH) has been fundamental to traditional AutoML systems. However, with the advancements of pre-trained models, modern ML workflows go beyond hyperparameter optimization and often require fine-tuning, ensembling, and other adaptation techniques. While the core challenge of identifying the best-performing model for a downstream task remains, the increasing heterogeneity of ML pipelines demands novel AutoML approaches. This work extends the CASH framework to select and adapt modern ML pipelines. We propose PS-PFN to efficiently explore and exploit adapting ML pipelines by extending Posterior Sampling (PS) to the max k -armed bandit problem setup. PS-PFN leverages prior-data fitted networks (PFNs) to efficiently estimate the posterior distribution of the maximal value via in-context learning. We show how to extend this method to consider varying costs of pulling arms and to use different PFNs to model reward distributions individually per arm. Experimental results on one novel and two existing standard benchmark tasks demonstrate the superior performance of PS-PFN compared to other bandit and AutoML strategies.

Document Visual Question Answering (Document VQA) faces significant challenges when processing long documents in low-resource environments due to context limitations and insufficient training data. This paper presents AdaDocVQA, a unified adaptive framework addressing these challenges through three core innovations: a hybrid text retrieval architecture for effective document segmentation, an intelligent data augmentation pipeline that automatically generates high-quality reasoning question-answer pairs with multi-level verification, and adaptive ensemble inference with dynamic configuration generation and early stopping mechanisms. Experiments on Japanese document VQA benchmarks demonstrate substantial improvements with 83.04\% accuracy on Yes/No questions, 52.66\% on factual questions, and 44.12\% on numerical questions in JDocQA, and 59\% accuracy on LAVA dataset. Ablation studies confirm meaningful contributions from each component, and our framework establishes new state-of-the-art results for Japanese document VQA while providing a scalable foundation for other low-resource languages and specialized domains. Our code available at: https://github.com/Haoxuanli-Thu/AdaDocVQA.

Improving large language models (LLMs) for electronic health record (EHR) reasoning is essential for enabling accurate and generalizable clinical predictions. While LLMs excel at medical text understanding, they underperform on EHR-based prediction tasks due to challenges in modeling temporally structured, high-dimensional data. Existing approaches often rely on hybrid paradigms, where LLMs serve merely as frozen prior retrievers while downstream deep learning (DL) models handle prediction, failing to improve the LLM's intrinsic reasoning capacity and inheriting the generalization limitations of DL models. To this end, we propose EAG-RL, a novel two-stage training framework designed to intrinsically enhance LLMs' EHR reasoning ability through expert attention guidance, where expert EHR models refer to task-specific DL models trained on EHR data. Concretely, EAG-RL first constructs high-quality, stepwise reasoning trajectories using expert-guided Monte Carlo Tree Search to effectively initialize the LLM's policy. Then, EAG-RL further optimizes the policy via reinforcement learning by aligning the LLM's attention with clinically salient features identified by expert EHR models. Extensive experiments on two real-world EHR datasets show that EAG-RL improves the intrinsic EHR reasoning ability of LLMs by an average of 14.62%, while also enhancing robustness to feature perturbations and generalization to unseen clinical domains. These results demonstrate the practical potential of EAG-RL for real-world deployment in clinical prediction tasks. Our code have been available at https://github.com/devilran6/EAG-RL.

Large Deep Learning models are often compressed before being deployed in a resource-constrained environment. Can we trust the prediction of compressed models just as we trust the prediction of the original large model? Existing work has keenly studied the effect of compression on accuracy and related performance measures. However, performance parity does not guarantee trust-equivalence. We propose a two-dimensional framework for trust-equivalence evaluation. First, interpretability alignment measures whether the models base their predictions on the same input features. We use LIME and SHAP tests to measure the interpretability alignment. Second, calibration similarity measures whether the models exhibit comparable reliability in their predicted probabilities. It is assessed via ECE, MCE, Brier Score, and reliability diagrams. We conducted experiments using BERT-base as the large model and its multiple compressed variants. We focused on two text classification tasks: natural language inference and paraphrase identification. Our results reveal low interpretability alignment and significant mismatch in calibration similarity. It happens even when the accuracies are nearly identical between models. These findings show that compressed models are not trust-equivalent to their large counterparts. Deploying compressed models as a drop-in replacement for large models requires careful assessment, going beyond performance parity.

Language model (LM) agents have demonstrated significant potential for automating real-world tasks, yet they pose a diverse array of potential, severe risks in safety-critical scenarios. In this work, we identify a significant gap between LM agents' risk awareness and safety execution abilities: while they often answer "Yes" to queries like "Is executing `sudo rm -rf /*' dangerous?", they will likely fail to identify such risks in instantiated trajectories or even directly perform these risky actions when acting as agents. To systematically investigate this, we develop a comprehensive evaluation framework to examine agents' safety across three progressive dimensions: 1) their knowledge about potential risks, 2) their ability to identify corresponding risks in execution trajectories, and 3) their actual behaviors to avoid executing these risky actions. Our evaluation reveals two critical performance gaps that resemble the generator-validator gaps observed in LMs: while agents demonstrate near-perfect risk knowledge ($>98\%$ pass rates), they fail to apply this knowledge when identifying risks in actual scenarios (with performance dropping by $>23\%$) and often still execute risky actions ($<26\%$ pass rates). Notably, this trend persists across more capable LMs as well as in specialized reasoning models like DeepSeek-R1, indicating that simply scaling model capabilities or inference compute does not inherently resolve safety concerns. Instead, we take advantage of these observed gaps to develop a risk verifier that independently critiques the proposed actions by agents, with an abstractor that converts specific execution trajectories into abstract descriptions where LMs can more effectively identify the risks. Our overall system achieves a significant reduction of risky action execution by $55.3\%$ over vanilla-prompted agents.

Recent advances in multimodal Parameter-Efficient Fine-Tuning (PEFT) have significantly improved performance on downstream tasks such as few-shot retrieval. However, most existing approaches focus on task-specific gains while neglecting the structure of the multimodal embedding space. As a result, modality-specific representations often remain isolated, limiting cross-modal generalisation. In this work, we introduce Shared Prompt AligNER (SP ANER), a modality-agnostic PEFT framework designed to embed inputs from diverse modalities into a unified semantic space. At its core, SP ANER employs a shared prompt mechanism that acts as a conceptual anchor, enabling semantically related instances to converge spatially regardless of modality. This shared prompt design is inherently extensible, supporting the seamless integration of additional modalities, such as audio, without altering the core architecture. Through comprehensive experiments across vision-language and audio-visual benchmarks, SP ANER demonstrates competitive few-shot retrieval performance while preserving high semantic coherence in the learned embedding space. Our results highlight the importance of aligning embedding structures, rather than merely tuning adapter weights, for scalable multimodal learning.

The recent trend towards utilisation of reasoning models has improved the performance of Large Language Models (LLMs) across many tasks which involve logical steps. One linguistic task that could benefit from this framing is idiomaticity detection, as a potentially idiomatic expression must first be understood before it can be disambiguated and serves as a basis for reasoning. In this paper, we explore how reasoning capabilities in LLMs affect idiomaticity detection performance and examine the effect of model size. We evaluate, as open source representative models, the suite of DeepSeek-R1 distillation models ranging from 1.5B to 70B parameters across four idiomaticity detection datasets. We find the effect of reasoning to be smaller and more varied than expected. For smaller models, producing chain-of-thought (CoT) reasoning increases performance from Math-tuned intermediate models, but not to the levels of the base models, whereas larger models (14B, 32B, and 70B) show modest improvements. Our in-depth analyses reveal that larger models demonstrate good understanding of idiomaticity, successfully producing accurate definitions of expressions, while smaller models often fail to output the actual meaning. For this reason, we also experiment with providing definitions in the prompts of smaller models, which we show can improve performance in some cases.

This paper tackles several challenges that arise when integrating Automatic Speech Recognition (ASR) and Machine Translation (MT) for real-time, on-device streaming speech translation. Although state-of-the-art ASR systems based on Recurrent Neural Network Transducers (RNN-T) can perform real-time transcription, achieving streaming translation in real-time remains a significant challenge. To address this issue, we propose a simultaneous translation approach that effectively balances translation quality and latency. We also investigate efficient integration of ASR and MT, leveraging linguistic cues generated by the ASR system to manage context and utilizing efficient beam-search pruning techniques such as time-out and forced finalization to maintain system's real-time factor. We apply our approach to an on-device bilingual conversational speech translation and demonstrate that our techniques outperform baselines in terms of latency and quality. Notably, our technique narrows the quality gap with non-streaming translation systems, paving the way for more accurate and efficient real-time speech translation.