Large Language Models (LLMs) are increasingly used as proxy students in the development of Intelligent Tutoring Systems (ITSs) and in piloting test questions. However, to what extent these proxy students accurately emulate the behavior and characteristics of real students remains an open question. To investigate this, we collected a dataset of 489 items from the National Assessment of Educational Progress (NAEP), covering mathematics and reading comprehension in grades 4, 8, and 12. We then apply an Item Response Theory (IRT) model to position 11 diverse and state-of-the-art LLMs on the same ability scale as real student populations. Our findings reveal that, without guidance, strong general-purpose models consistently outperform the average student at every grade, while weaker or domain-mismatched models may align incidentally. Using grade-enforcement prompts changes models' performance, but whether they align with the average grade-level student remains highly model- and prompt-specific: no evaluated model-prompt pair fits the bill across subjects and grades, underscoring the need for new training and evaluation strategies. We conclude by providing guidelines for the selection of viable proxies based on our findings.

We introduce the LLM-Nash framework, a game-theoretic model where agents select reasoning prompts to guide decision-making via Large Language Models (LLMs). Unlike classical games that assume utility-maximizing agents with full rationality, this framework captures bounded rationality by modeling the reasoning process explicitly. Equilibrium is defined over the prompt space, with actions emerging as the behavioral output of LLM inference. This approach enables the study of cognitive constraints, mindset expressiveness, and epistemic learning. Through illustrative examples, we show how reasoning equilibria can diverge from classical Nash outcomes, offering a new foundation for strategic interaction in LLM-enabled systems.

As large language models (LLMs) are increasingly deployed in critical applications, the challenge of jailbreaking, where adversaries manipulate the models to bypass safety mechanisms, has become a significant concern. This paper presents a dynamic Stackelberg game framework to model the interactions between attackers and defenders in the context of LLM jailbreaking. The framework treats the prompt-response dynamics as a sequential extensive-form game, where the defender, as the leader, commits to a strategy while anticipating the attacker's optimal responses. We propose a novel agentic AI solution, the "Purple Agent," which integrates adversarial exploration and defensive strategies using Rapidly-exploring Random Trees (RRT). The Purple Agent actively simulates potential attack trajectories and intervenes proactively to prevent harmful outputs. This approach offers a principled method for analyzing adversarial dynamics and provides a foundation for mitigating the risk of jailbreaking.

Generative Large Language Models (LLMs)inevitably produce untruthful responses. Accurately predicting the truthfulness of these outputs is critical, especially in high-stakes settings. To accelerate research in this domain and make truthfulness prediction methods more accessible, we introduce TruthTorchLM an open-source, comprehensive Python library featuring over 30 truthfulness prediction methods, which we refer to as Truth Methods. Unlike existing toolkits such as Guardrails, which focus solely on document-grounded verification, or LM-Polygraph, which is limited to uncertainty-based methods, TruthTorchLM offers a broad and extensible collection of techniques. These methods span diverse tradeoffs in computational cost, access level (e.g., black-box vs white-box), grounding document requirements, and supervision type (self-supervised or supervised). TruthTorchLM is seamlessly compatible with both HuggingFace and LiteLLM, enabling support for locally hosted and API-based models. It also provides a unified interface for generation, evaluation, calibration, and long-form truthfulness prediction, along with a flexible framework for extending the library with new methods. We conduct an evaluation of representative truth methods on three datasets, TriviaQA, GSM8K, and FactScore-Bio. The code is available at https://github.com/Ybakman/TruthTorchLM

The emergence of large language models (LLMs) and agentic systems is enabling autonomous 6G networks with advanced intelligence, including self-configuration, self-optimization, and self-healing. However, the current implementation of individual intelligence tasks necessitates isolated knowledge retrieval pipelines, resulting in redundant data flows and inconsistent interpretations. Inspired by the service model unification effort in Open-RAN (to support interoperability and vendor diversity), we propose KP-A: a unified Network Knowledge Plane specifically designed for Agentic network intelligence. By decoupling network knowledge acquisition and management from intelligence logic, KP-A streamlines development and reduces maintenance complexity for intelligence engineers. By offering an intuitive and consistent knowledge interface, KP-A also enhances interoperability for the network intelligence agents. We demonstrate KP-A in two representative intelligence tasks: live network knowledge Q&A and edge AI service orchestration. All implementation artifacts have been open-sourced to support reproducibility and future standardization efforts.

Machine learning has been proposed as a way to improve educational assessment by making fine-grained predictions about student performance and learning relationships between items. One challenge with many machine learning approaches is incorporating new items, as these approaches rely heavily on historical data. We develop Text-LENS by extending the LENS partial variational auto-encoder for educational assessment to leverage item text embeddings, and explore the impact on predictive performance and generalization to previously unseen items. We examine performance on two datasets: Eedi, a publicly available dataset that includes item content, and LLM-Sim, a novel dataset with test items produced by an LLM. We find that Text-LENS matches LENS' performance on seen items and improves upon it in a variety of conditions involving unseen items; it effectively learns student proficiency from and makes predictions about student performance on new items.

The distillation of knowledge from Large Language Models (LLMs) into Smaller Language Models (SLMs), preserving the capabilities and performance of LLMs while reducing model size, has played a key role in the proliferation of LLMs. Because SLMs are considerably smaller than LLMs, they are often utilized in domains where human interaction is frequent but resources are highly constrained, e.g., smart phones. Therefore, it is crucial to ensure that empathy, a fundamental aspect of positive human interactions, already instilled into LLMs, is retained by SLMs after distillation. In this paper, we develop a comprehensive approach for effective empathy distillation from LLMs into SLMs. Our approach features a two-step fine-tuning process that fully leverages datasets of empathetic dialogue responses distilled from LLMs. We explore several distillation methods beyond basic direct prompting and propose four unique sets of prompts for targeted empathy improvement to significantly enhance the empathy distillation process. Our evaluations demonstrate that SLMs fine-tuned through the two-step fine-tuning process with distillation datasets enhanced by the targeted empathy improvement prompts significantly outperform the base SLM at generating empathetic responses with a win rate of 90%. Our targeted empathy improvement prompts substantially outperform the basic direct prompting with a 10% improvement in win rate.

Contemporary organizations have shown great interest in integrating language models (LMs) into workflows traditionally performed by human subject matter experts (SMEs), such as in medical diagnostics (Artsi et al., 2025), legal assistance (Padiu et al., 2024), financial risk analysis (AI21 labs, 2025), and governmental permitting or regulatory reviews (Phan et al., 2024). Despite this interest, however, the use of LMs (e.g., via a standard conversational interface) in high-stakes contexts is constrained by the need for decision-making reliability, objectivity, transparency, and accountability that SMEs currently provide (Mori, 2024). Effective reconciliation between LMs and SMEs thus represents a critical frontier in real-world deployments of artificial intelligence. LMs have demonstrated remarkable capabilities in extracting information from large volumes of multi-modal, multi-domain data; synthesizing multi-document concepts; and performing tasks associated with basic reasoning. Nonetheless, LMs are susceptible to "hallucinations" (i.e., inaccurate generation) (Ji et al., 2023), difficulty in handling nuanced, domain-specific requirements (Ashqar, 2025), historical biases inherited from training data (Ranjan et al., 2024), and opaque reasoning in decision-making (Machot et al., 2024). Notably, these weaknesses are often precisely the strengths of SMEs, who are conversely burdened with the inefficient and labor-intensive tasks of cross-document, multi-modal search and information extraction. We can see the need to delineate and integrate the often low-stakes or tedious work that can be performed by LMs with the discerning, high-stakes decision-making tasks performed by SMEs in the real world: The challenge is to harness the time efficiency and broad knowledge capabilities of LMs while preserving the domain expertise, contextual judgment, oversight, and accountability of SMEs. Moreover, we must do so without creating additional burdens for SMEs to work with LMs (e.g., "prompt-engineering" or manual review of all LM tasks), and we wish to minimize the introduction of new risks (e.g., a loss of clarity regarding where or how LMs may be used by each SME, or, in the case of governmental work, the erosion of public trust). In this work, we propose a novel auditable and interactive refinement framework for the effective integration of LMs with SMEs for decision-making workflows.

Social media has amplified the reach of financial influencers known as "finfluencers," who share stock recommendations on platforms like YouTube. Understanding their influence requires analyzing multimodal signals like tone, delivery style, and facial expressions, which extend beyond text-based financial analysis. We introduce VideoConviction, a multimodal dataset with 6,000+ expert annotations, produced through 457 hours of human effort, to benchmark multimodal large language models (MLLMs) and text-based large language models (LLMs) in financial discourse. Our results show that while multimodal inputs improve stock ticker extraction (e.g., extracting Apple's ticker AAPL), both MLLMs and LLMs struggle to distinguish investment actions and conviction--the strength of belief conveyed through confident delivery and detailed reasoning--often misclassifying general commentary as definitive recommendations. While high-conviction recommendations perform better than low-conviction ones, they still underperform the popular S\&P 500 index fund. An inverse strategy--betting against finfluencer recommendations--outperforms the S\&P 500 by 6.8\% in annual returns but carries greater risk (Sharpe ratio of 0.41 vs. 0.65). Our benchmark enables a diverse evaluation of multimodal tasks, comparing model performance on both full video and segmented video inputs. This enables deeper advancements in multimodal financial research. Our code, dataset, and evaluation leaderboard are available under the CC BY-NC 4.0 license.

Fine-tuning large foundation models presents significant memory challenges due to stateful optimizers like AdamW, often requiring several times more GPU memory than inference. While memory-efficient methods like parameter-efficient fine-tuning (e.g., LoRA) and optimizer state compression exist, recent approaches like GaLore bridge these by using low-rank gradient projections and subspace moment accumulation. However, such methods may struggle with fixed subspaces or computationally costly offline resampling (e.g., requiring full-matrix SVDs). We propose Momentum Factorized SGD (MoFaSGD), which maintains a dynamically updated low-rank SVD representation of the first-order momentum, closely approximating its full-rank counterpart throughout training. This factorization enables a memory-efficient fine-tuning method that adaptively updates the optimization subspace at each iteration. Crucially, MoFaSGD leverages the computed low-rank momentum factors to perform efficient spectrally normalized updates, offering an alternative to subspace moment accumulation. We establish theoretical convergence guarantees for MoFaSGD, proving it achieves an optimal rate for non-convex stochastic optimization under standard assumptions. Empirically, we demonstrate MoFaSGD's effectiveness on large language model alignment benchmarks, achieving a competitive trade-off between memory reduction (comparable to LoRA) and performance compared to state-of-the-art low-rank optimization methods.