Foundation Models that are capable of processing and generating multi-modal data have transformed AI's role in medicine. However, a key limitation of their reliability is hallucination, where inaccurate or fabricated information can impact clinical decisions and patient safety. We define medical hallucination as any instance in which a model generates misleading medical content. This paper examines the unique characteristics, causes, and implications of medical hallucinations, with a particular focus on how these errors manifest themselves in real-world clinical scenarios. Our contributions include (1) a taxonomy for understanding and addressing medical hallucinations, (2) benchmarking models using medical hallucination dataset and physician-annotated LLM responses to real medical cases, providing direct insight into the clinical impact of hallucinations, and (3) a multi-national clinician survey on their experiences with medical hallucinations. Our results reveal that inference techniques such as Chain-of-Thought (CoT) and Search Augmented Generation can effectively reduce hallucination rates. However, despite these improvements, non-trivial levels of hallucination persist. These findings underscore the ethical and practical imperative for robust detection and mitigation strategies, establishing a foundation for regulatory policies that prioritize patient safety and maintain clinical integrity as AI becomes more integrated into healthcare. The feedback from clinicians highlights the urgent need for not only technical advances but also for clearer ethical and regulatory guidelines to ensure patient safety. A repository organizing the paper resources, summaries, and additional information is available at https://github.com/mitmedialab/medical hallucination.

AI systems powered by large language models can act as capable assistants for writing and editing. In these tasks, the AI system acts as a co-creative partner, making novel contributions to an artifact-under-creation alongside its human partner(s). One question that arises in these scenarios is the extent to which AI should be credited for its contributions. We examined knowledge workers' views of attribution through a survey study (N=155) and found that they assigned different levels of credit across different contribution types, amounts, and initiative. Compared to a human partner, we observed a consistent pattern in which AI was assigned less credit for equivalent contributions. Participants felt that disclosing AI involvement was important and used a variety of criteria to make attribution judgments, including the quality of contributions, personal values, and technology considerations. Our results motivate and inform new approaches for crediting AI contributions to co-created work.

We study the asymptotic spectral behavior of the conjugate kernel random matrix $YY^\top$, where $Y= f(WX)$ arises from a two-layer neural network model. We consider the setting where $W$ and $X$ are both random rectangular matrices with i.i.d. entries, where the entries of $W$ follow a heavy-tailed distribution, while those of $X$ have light tails. Our assumptions on $W$ include a broad class of heavy-tailed distributions, such as symmetric $\alpha$-stable laws with $\alpha \in (0,2)$ and sparse matrices with $\mathcal{O}(1)$ nonzero entries per row. The activation function $f$, applied entrywise, is nonlinear, smooth, and odd. By computing the eigenvalue distribution of $YY^\top$ through its moments, we show that heavy-tailed weights induce strong correlations between the entries of $Y$, leading to richer and fundamentally different spectral behavior compared to models with light-tailed weights.

The scalability of instructable agents in robotics or gaming is often hindered by limited data that pairs instructions with agent trajectories. However, large datasets of unannotated trajectories containing sequences of various agent behaviour (play trajectories) are often available. In a semi-supervised setup, we explore methods to extract labelled segments from play trajectories. The goal is to augment a small annotated dataset of instruction-trajectory pairs to improve the performance of an instruction-following policy trained downstream via imitation learning. Assuming little variation in segment length, recent video segmentation methods can effectively extract labelled segments. To address the constraint of segment length, we propose Play Segmentation (PS), a probabilistic model that finds maximum likely segmentations of extended subsegments, while only being trained on individual instruction segments. Our results in a game environment and a simulated robotic gripper setting underscore the importance of segmentation; randomly sampled segments diminish performance, while incorporating labelled segments from PS improves policy performance to the level of a policy trained on twice the amount of labelled data.

Preference learning is critical for aligning large language models (LLMs) with human values, with the quality of preference datasets playing a crucial role in this process. While existing metrics primarily assess data quality based on either explicit or implicit reward margins, they often provide contradictory evaluations for the same data. To address this issue, we introduce the alignment potential metric, which quantifies the gap from the model's current implicit reward margin to the target explicit reward margin, thereby estimating the model's potential to align with the preference data. Empirical results demonstrate that training on data selected by this metric consistently enhances alignment performance, surpassing existing metrics across different base models and optimization objectives. Furthermore, our method extends to self-play data generation frameworks, where the metric is used to identify high-quality data within the self-generated content by LLMs. Under this data generation scenario, our method surpasses current state-of-the-art (SOTA) results across various training settings and demonstrates continuous improvements in alignment performance as dataset size and training iterations increase.

Evaluating creativity is challenging, even for humans, not only because of its subjectivity but also because it involves complex cognitive processes. Inspired by work in marketing, we attempt to break down visual advertisement creativity into atypicality and originality. With fine-grained human annotations on these dimensions, we propose a suit of tasks specifically for such a subjective problem. We also evaluate the alignment between state-of-the-art (SoTA) vision language models (VLM) and humans on our proposed benchmark, demonstrating both the promises and challenges of using VLMs for automatic creativity assessment.

Early detection of suicide risk from social media text is crucial for timely intervention. While Large Language Models (LLMs) offer promising capabilities in this domain, challenges remain in terms of interpretability and computational efficiency. This paper introduces Evidence-Driven LLM (ED-LLM), a novel approach for clinical marker extraction and suicide risk classification. ED-LLM employs a multi-task learning framework, jointly training a Mistral-7B based model to identify clinical marker spans and classify suicide risk levels. This evidence-driven strategy enhances interpretability by explicitly highlighting textual evidence supporting risk assessments. Evaluated on the CLPsych datasets, ED-LLM demonstrates competitive performance in risk classification and superior capability in clinical marker span identification compared to baselines including fine-tuned LLMs, traditional machine learning, and prompt-based methods. The results highlight the effectiveness of multi-task learning for interpretable and efficient LLM-based suicide risk assessment, paving the way for clinically relevant applications.

Social surveys in computational social science are well-designed by elaborate domain theories that can effectively reflect the interviewee's deep thoughts without concealing their true feelings. The candidate questionnaire options highly depend on the interviewee's previous answer, which results in the complexity of social survey analysis, the time, and the expertise required. The ability of large language models (LLMs) to perform complex reasoning is well-enhanced by prompting learning such as Chain-of-thought (CoT) but still confined to left-to-right decision-making processes or limited paths during inference. This means they can fall short in problems that require exploration and uncertainty searching. In response, a novel large language model prompting method, called Random Forest of Thoughts (RFoT), is proposed for generating uncertainty reasoning to fit the area of computational social science. The RFoT allows LLMs to perform deliberate decision-making by generating diverse thought space and randomly selecting the sub-thoughts to build the forest of thoughts. It can extend the exploration and prediction of overall performance, benefiting from the extensive research space of response. The method is applied to optimize computational social science analysis on two datasets covering a spectrum of social survey analysis problems. Our experiments show that RFoT significantly enhances language models' abilities on two novel social survey analysis problems requiring non-trivial reasoning.

Using a sample of 25,000 Bing Copilot conversations, we study how the agent responds to users of varying levels of domain expertise and the resulting impact on user experience along multiple dimensions. Our findings show that across a variety of topical domains, the agent largely responds at proficient or expert levels of expertise (77% of conversations) which correlates with positive user experience regardless of the user's level of expertise. Misalignment, such that the agent responds at a level of expertise below that of the user, has a negative impact on overall user experience, with the impact more profound for more complex tasks. We also show that users engage more, as measured by the number of words in the conversation, when the agent responds at a level of expertise commensurate with that of the user. Our findings underscore the importance of alignment between user and AI when designing human-centered AI systems, to ensure satisfactory and productive interactions.

Knowledge components (KCs) mapped to problems help model student learning, tracking their mastery levels on fine-grained skills thereby facilitating personalized learning and feedback in online learning platforms. However, crafting and tagging KCs to problems, traditionally performed by human domain experts, is highly labor-intensive. We present a fully automated, LLM-based pipeline for KC generation and tagging for open-ended programming problems. We also develop an LLM-based knowledge tracing (KT) framework to leverage these LLM-generated KCs, which we refer to as KCGen-KT. We conduct extensive quantitative and qualitative evaluations validating the effectiveness of KCGen-KT. On a real-world dataset of student code submissions to open-ended programming problems, KCGen-KT outperforms existing KT methods. We investigate the learning curves of generated KCs and show that LLM-generated KCs have a comparable level-of-fit to human-written KCs under the performance factor analysis (PFA) model. We also conduct a human evaluation to show that the KC tagging accuracy of our pipeline is reasonably accurate when compared to that by human domain experts.