artificial intelligence

Towards Portable Specification of the Agent Harness

The harness is decompose into eight functional components that can be specified and compared across systems, and a portable specification framework is proposed that records which model ran in which harness.

Does Recursive Knowledge Construction Help Agents Verify Consensus Claims in Science?

It is found that the evidence state improves performance over in-context accumulation, and it also matters what is stored in it.

ChatGPT as a Surveillance Assistant? An Exploratory Case Study of AI-Assisted Surveillance Risks

An exploratory case study of what potentially harmful surveillance assistance looks like in ChatGPT's public web interface and finds that even when ChatGPT states a refusal, it may still supply detailed instructions, working code, or architectural guidance alongside brief ethical caveats.

Incongruence-Aware Evaluation of Multimodal Speech Emotion Recognition in Audio Large Language Models

Mapping text and audio into a shared valence-arousal-dominance space, a continuous, threshold-free incongruence score is converted into a continuous, threshold-free incongruence score with a two-component Gaussian mixture, and validated against human judgement.

Circumventing the Computational Hardness of Computational Arbitrage

This work formalizes computational arbitrage as an allocation problem and proves that it is generally NP-hard, but shows that the optimal allocation can be computed efficiently under a natural diminishing-returns condition on how model performance scales with inference spending.

Beyond Plausible Prediction: Safety Sufficiency in Web World Models

SafeWebWorld is introduced, a benchmark for evaluating safety sufficiency across five web-safety domains through action-conditioned futures, policy-decisive evidence, and policy-conditioned decisions, and it is shown that aggregate recoverability does not imply robust fine-grained authorisation behaviour.

More Rules, Longer Lists? Benchmarking Self-Evolving Agents on a Hidden-Rule Curriculum

This work evaluates six recent self-evolution methods and introduces CLR-Bench, a Curriculum of Latent Rules benchmark, which measures how well an agent learns, and this ability differs across models.

Beyond public access in LLM pre-training data

The results suggest that GPT-4o, OpenAI’s more recent and capable model in the authors' sample, exhibits patterns consistent with recognition of non-public books, and support disclosure requirements for pre-training data provenance and formal licensing markets for training content.

SprintX: mitigating data scarcity in antibody design via synthetic structural prior knowledge dataset.

Across downstream tasks including antibody generation, computational binding-affinity prediction, and binding-interface prediction, models pretrained with SprintX showed meaningful improvements in structural plausibility and competitive performance on computational binding-affinity and binding-interface prediction benc...

TOURISM 5.0 IN PRACTICE: NAVIGATING THE TENSIONS BETWEEN DIGITAL EFFICIENCY AND LIVED AUTHENTICITY IN THE SAUDI CONTEXT

The Tourism 5.0 paradigm is emerging as a pivotal response to the limits of a purely efficiency-oriented digitalisation, by placing people and sustainability back at the centre of innovation to create a holistic and responsible tourism ecosystem. Within the framework of Vision 2030, Saudi Arabia has adopted this orient...

Artificial intelligence and machine learning models for electrolyzers and fuel cells optimization: A review on empowering sustainable hydrogen economy

Never-Skilling in Dentistry: When Artificial Intelligence Performs the Thinking Before Learners Learn to Think