探索全球前沿学术脉络

Agents' Last Exam

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long horizon, economically valuable, real world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 sub fields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is below 1%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP relevant impact.

NeST: Neuron Selective Tuning for LLM Safety

arXiv:2602.16835v2 Announce Type: replace-cross Abstract: Safety alignment is essential for the responsible deployment of Large Language Models (LLMs). Yet, existing approaches often rely on heavyweight fine-tuning that is costly to update, audit, and maintain across model families. Full fine-tuning incurs substantial computational and storage overhead, while parameter-efficient methods, e.g., Low-Rank Adaptation (LoRA), trade efficiency for inconsistent safety gains and sensitivity to design choices. Safety intervention mechanisms reduce unsafe outputs without modifying model weights, but do not directly shape or preserve the internal representations that govern safety behavior. We present NeST, a Neuron-Selective Tuning framework for efficient post-hoc safety alignment. NeST identifies safety-relevant feed-forward neurons via activation probing on vanilla harmful and benign prompts, clusters neurons with similar activation profiles, and trains shared cluster-level updates while freezing the rest of the model. Importantly, NeST is trained only on vanilla malicious prompts, without using jailbreak-specific attack data, yet generalizes robustly to diverse jailbreaks. The learned updates are then folded into the original weights, incurring no inference-time overhead. Evaluated on 14 open-weight language and multimodal models, NeST outperforms lightweight baselines and approaches full fine-tuning robustness with significantly fewer trainable parameters. On text-only models, NeST reduces average jailbreak attack success rate from 44.5% to 1.1% while training only 0.4M parameters on average. Across multimodal settings, it reduces ASR from 55.3% to 1.1%, and for downstream fine-tuned variants, it restores safety by reducing ASR from 53.8% to 0.8%. These results show that robust, maintainable safety alignment can be achieved by concentrating adaptation on localized, functionally coherent safety structures.

UniMM: A Unified Mixture Model Framework for Multi-Agent Simulation

arXiv:2501.17015v2 Announce Type: replace Abstract: Simulation plays a crucial role in assessing autonomous driving systems, where the generation of realistic multi-agent behaviors is a key aspect. In multi-agent simulation, the primary challenges include behavioral multimodality and closed-loop distributional shifts. In this study, we formulate a unified mixture model (UniMM) framework for generating multimodal agent behaviors, which can cover the mainstream methods including regression-based mixture models and discrete NTP models. Furthermore, we introduce a closed-loop sample generation approach tailored for mixture models to mitigate distributional shifts. Within the UniMM framework, we recognize critical configurations from both the model and data perspectives. We conduct a systematic examination of various model configurations, and comprehensively characterize their effects. Moreover, our investigation into the data configuration highlights the pivotal role of closed-loop samples in achieving realistic simulations. To extend the benefits of closed-loop samples across a broader range of mixture models, we further introduce a temporal disentanglement-and-alignment mechanism to address the shortcut learning and off-policy learning issues. Leveraging insights from our exploration, the distinct variants proposed within the UniMM framework, including discrete, anchor-free, and anchor-based models, all achieve state-of-the-art performance on the WOSAC benchmark.

SPARK: Security Knowledge Priming and Representation-Guided Knowledge Activation for LLM-based Secure Code Generation

arXiv:2606.16244v1 Announce Type: cross Abstract: Large language models routinely generate code with exploitable security flaws. Prior literature attributes this limitation to a lack of security expertise, steering current defense mechanisms toward heavy fine-tuning or external knowledge retrieval, which introduces significant computational overhead and data bias through redundant code examples. Contrary to this view, we argue that pretraining corpora are already rich in security material. The bottleneck is activation: without an explicit and brief cue, statistical pressure toward common training-distribution patterns suppresses the model's safety-relevant representations. We present SPARK, an inference-time security harness that activates this latent knowledge without any retraining. The harness has two parts. Component~I retrieves a few of the relevant Common Weakness Enumeration (CWE) entries for each coding task and appends a short structured cue to the prompt; this alone is enough to surface the model's existing security representations. Component~II adds a precomputed token bias to the logits at every decoding step. We obtain the bias by projecting a safe-direction vector, the unit difference between the mean safe and mean unsafe last-layer hidden states, through the language model head. The bias is computed once offline; applying it costs a single vector addition per generated token. We evaluate SPARK on 9 open-source models across C++, Java, and Python, and compare with 7 baselines spanning fine-tuning and retrieval-augmented methods. SPARK matches or improves on the best baseline in every setting while preserving HumanEval utility. We further test Component~I in a black-box setting on 7 of today's strongest models, including Claude, DeepSeek, and GPT, demonstrating the bottleneck of insecure code generation and the improvements enabled by our method.