751 papers
This empirical study introduces a "Lie Awareness Spectrum" to characterize how GPT-5.4, Claude 4.6 Sonnet, and Gemini 3.1 Pro differentially recognize, rationalize, and refuse deception, revealing distinct ethical architectures and a paradox where models resist direct lying commands yet succumb to implicit social pressure.
This study demonstrates that AI tools can rapidly retrieve relevant medical literature and effectively detect significant deviations from evidence in expert testimony within medical malpractice cases, particularly highlighting that defense experts often contradict published data more frequently than plaintiff experts.
This study demonstrates that while sentiment intensity, readability, and lexical diversity reliably distinguish positive from negative reviews across film and consumer-electronics domains, affective punctuation exhibits contradictory valence signals between categories, highlighting the critical need for domain-specific calibration in automated sentiment analysis systems.
This study evaluates the temporal reliability of four successive ChatGPT versions in bilingual translation across finance, technology, and public policy domains, revealing that model updates do not guarantee consistent improvements in semantic accuracy or readability and instead exhibit diverging performance trajectories that necessitate continuous, domain-specific evaluations for AI-assisted language applications.
This study demonstrates that for wildlife re-identification with small datasets, using a frozen large foundation model like DINOv2-Giant to extract features and train a lightweight metric-learning head can outperform fine-tuned specialized models, offering a reproducible, low-overhead framework for conservationists working with pattern-bearing species.
This paper addresses the critical lack of computational resources for Myanmar's ancient Pyu script by introducing the first publicly available, systematically curated handwritten dataset of 33 consonants and validating a modified ResNet-18 architecture as a foundational benchmark for future digital preservation and epigraphic research.
DERMA-Agent is a novel agentic framework that leverages multimodal foundation models and dynamic statistical execution to autonomously generate and validate morphomolecular prognostic signatures from pan-cancer whole slide images, addressing the limitations of static supervised models through iterative hypothesis discovery while maintaining rigorous safety controls in a retrospective research setting.
This paper presents a privacy-preserving keystroke dynamics authentication system implemented using the CKKS homomorphic encryption scheme, which enables secure template matching via encrypted squared Euclidean distance calculations while ensuring biometric vectors remain confidential throughout the process.
This paper proposes a continual and open-ended framework that learns to schedule autonomous robots' thinking activities based on historical data and environmental context, significantly improving task success rates and cross-task generalization while reducing redundant LLM calls and operational violations.
This paper demonstrates that the sharp "emergent" jumps in exact-match accuracy observed in multi-digit integer addition are not genuine discontinuities in model capability but are instead fully predictable artifacts resulting from smooth, power-law improvements in per-token accuracy compounded by the nonlinearity of requiring all digits to be correct.