Large Language Models (LLMs) have demonstrated impressive capabilities in generating coherent text but often struggle with grounding language and strategic dialogue. To address this gap, we focus on journalistic interviews, a domain rich in grounding communication and abundant in data. We curate a dataset of 40,000 two-person informational interviews from NPR and CNN, and reveal that LLMs are significantly less likely than human interviewers to use acknowledgements and to pivot to higher-level questions. Realizing that a fundamental deficit exists in multi-turn planning and strategic thinking, we develop a realistic simulated environment, incorporating source personas and persuasive elements, in order to facilitate the development of agents with longer-horizon rewards. Our experiments show that while source LLMs mimic human behavior in information sharing, interviewer LLMs struggle with recognizing when questions are answered and engaging persuasively, leading to suboptimal information extraction across model size and capability. These findings underscore the need for enhancing LLMs’ strategic dialogue capabilities.

Large Language Models (LLMs) are primarily trained on high-resource natural languages, limiting their effectiveness in low-resource settings and in tasks requiring deep logical reasoning. This research introduces Rosetta-PL, a benchmark designed to evaluate LLMs’ logical reasoning and generalization capabilities in a controlled environment. We construct Rosetta-PL by translating a dataset of logical propositions from Lean into a custom logical language, which is then used to fine-tune an LLM (e.g., GPT-4o). Our experiments analyze the impact of the size of the dataset and the translation methodology on the performance of the model. Our results indicate that preserving logical relationships in the translation process significantly boosts precision, with accuracy plateauing beyond roughly 20,000 training samples. These insights provide valuable guidelines for optimizing LLM training in formal reasoning tasks and improving performance in various low-resource language applications.

Deception detection is crucial in domains such as security, forensics, and legal proceedings, as well as to ensure the reliability of AI systems. However, current approaches are limited by the lack of generalizable and interpretable benchmarks built on large and diverse datasets. To address this gap, we introduce DecepBench, a comprehensive and robust benchmark for multimodal deception detection. DecepBench includes an enhanced version of the DOLOS dataset, the largest game-show deception dataset (1,700 labeled video clips with audio). We augment each video clip with transcripts, introducing a third modality (text) and incorporating deception-related features identified in psychological research. We employ explainable methods to evaluate the relevance of key deception cues, providing insights into model limitations and guiding future improvements. Our enhancements to DOLOS, combined with these interpretable analyses, yield improved performance and a deeper understanding of multimodal deception detection.