This report synthesizes the outcomes of a recent interdisciplinary workshop that brought together leading experts in cognitive psychology, language learning, and artificial intelligence (AI)-based natural language processing (NLP). The workshop, funded by the National Science Foundation, aimed to address a critical knowledge gap in our understanding of the relationship between AI language models and human cognitive processes in text comprehension and composition. Through collaborative dialogue across cognitive, linguistic, and technological perspectives, workshop participants examined the underlying processes involved when humans produce and comprehend text, and how AI can both inform our understanding of these processes and augment human capabilities. The workshop revealed emerging patterns in the relationship between large language models (LLMs) and human cognition, with highlights on both the capabilities of LLMs and their limitations in fully replicating human-like language understanding and generation. Key findings include the potential of LLMs to offer insights into human language processing, the increasing alignment between LLM behavior and human language processing when models are fine-tuned with human feedback, and the opportunities and challenges presented by human-AI collaboration in language tasks. By synthesizing these findings, this report aims to guide future research, development, and implementation of LLMs in cognitive psychology, linguistics, and education. It emphasizes the importance of ethical considerations and responsible use of AI technologies while striving to enhance human capabilities in text comprehension and production through effective human-AI collaboration.

Benchmarks like Massive Multitask Language Understanding (MMLU) have played a pivotal role in evaluating AI's knowledge and abilities across diverse domains. However, existing benchmarks predominantly focus on content knowledge, leaving a critical gap in assessing models' understanding of pedagogy - the method and practice of teaching. This paper introduces The Pedagogy Benchmark, a novel dataset designed to evaluate large language models on their Cross-Domain Pedagogical Knowledge (CDPK) and Special Education Needs and Disability (SEND) pedagogical knowledge. These benchmarks are built on a carefully curated set of questions sourced from professional development exams for teachers, which cover a range of pedagogical subdomains such as teaching strategies and assessment methods. Here we outline the methodology and development of these benchmarks. We report results for 97 models, with accuracies spanning a range from 28% to 89% on the pedagogical knowledge questions. We consider the relationship between cost and accuracy and chart the progression of the Pareto value frontier over time. We provide online leaderboards at https://rebrand.ly/pedagogy which are updated with new models and allow interactive exploration and filtering based on various model properties, such as cost per token and open-vs-closed weights, as well as looking at performance in different subjects. LLMs and generative AI have tremendous potential to influence education and help to address the global learning crisis. Education-focused benchmarks are crucial to measure models' capacities to understand pedagogical concepts, respond appropriately to learners' needs, and support effective teaching practices across diverse contexts. They are needed for informing the responsible and evidence-based deployment of LLMs and LLM-based tools in educational settings, and for guiding both development and policy decisions.

Scientific information expresses human understanding of nature. This knowledge is largely disseminated in different forms of text, including scientific papers, news articles, and discourse among people on social media. While important for accelerating our pursuit of knowledge, not all scientific text is faithful to the underlying science. As the volume of this text has burgeoned online in recent years, it has become a problem of societal importance to be able to identify the faithfulness of a given piece of scientific text automatically. This thesis is concerned with the cultivation of datasets, methods, and tools for machine understanding of scientific language, in order to analyze and understand science communication at scale. To arrive at this, I present several contributions in three areas of natural language processing and machine learning: automatic fact checking, learning with limited data, and scientific text processing. These contributions include new methods and resources for identifying check-worthy claims, adversarial claim generation, multi-source domain adaptation, learning from crowd-sourced labels, cite-worthiness detection, zero-shot scientific fact checking, detecting exaggerated scientific claims, and modeling degrees of information change in science communication. Critically, I demonstrate how the research outputs of this thesis are useful for effectively learning from limited amounts of scientific text in order to identify misinformative scientific statements and generate new insights into the science communication process

Recent advances in machine learning, particularly the emergence of foundation models, are leading to new opportunities to develop technology-based solutions to societal problems. However, the reasoning and inner workings of today's complex AI models are not transparent to the user, and there are no safety guarantees regarding their predictions. Consequently, to fulfill the promise of AI, we must address the following scientific challenge: how to develop AI-based systems that are not only accurate and performant but also safe and trustworthy? The criticality of safe operation is particularly evident for autonomous systems for control and robotics, and was the catalyst for the Safe Learning Enabled Systems (SLES) program at NSF. For the broader class of AI applications, such as users interacting with chatbots and clinicians receiving treatment recommendations, safety is, while no less important, less well-defined with context-dependent interpretations. This motivated the organization of a day-long workshop, held at University of Pennsylvania on February 26, 2025, to bring together investigators funded by the NSF SLES program with a broader pool of researchers studying AI safety. This report is the result of the discussions in the working groups that addressed different aspects of safety at the workshop. The report articulates a new research agenda focused on developing theory, methods, and tools that will provide the foundations of the next generation of AI-enabled systems.

We argue that viewing AI as a simple reduction in input costs overlooks two critical dynamics: (a) the inherent trade-offs among generality, accuracy, and simplicity, and (b) the redistribution of complexity across stakeholders. While LLMs appear to defy the traditional trade-off by offering high generality and accuracy through simple interfaces, this user-facing simplicity masks a significant shift of complexity to infrastructure, compliance, and specialized personnel. The GAS trade-off, therefore, does not disappear but is relocated from the user to the organization, creating new managerial challenges, particularly around accuracy in high-stakes applications. We contend that competitive advantage no longer stems from mere AI adoption, but from mastering this redistributed complexity through the design of abstraction layers, workflow alignment, and complementary expertise.

With the proliferation of large language model (LLM) applications since 2022, their use in education has sparked both excitement and concern. Recent studies consistently highlight students' (mis)use of LLMs can hinder learning outcomes. This work aims to teach students how to effectively prompt LLMs to improve their learning. We first proposed pedagogical prompting, a theoretically-grounded new concept to elicit learning-oriented responses from LLMs. To move from concept design to a proof-of-concept learning intervention in real educational settings, we selected early undergraduate CS education (CS1/CS2) as the example context. We began with a formative survey study with instructors (N=36) teaching early-stage undergraduate-level CS courses to inform the instructional design based on classroom needs. Based on their insights, we designed and developed a learning intervention through an interactive system with scenario-based instruction to train pedagogical prompting skills. Finally, we evaluated its instructional effectiveness through a user study with CS novice students (N=22) using pre/post-tests. Through mixed methods analyses, our results indicate significant improvements in learners' LLM-based pedagogical help-seeking skills, along with positive attitudes toward the system and increased willingness to use pedagogical prompts in the future. Our contributions include (1) a theoretical framework of pedagogical prompting; (2) empirical insights into current instructor attitudes toward pedagogical prompting; and (3) a learning intervention design with an interactive learning tool and scenario-based instruction leading to promising results on teaching LLM-based help-seeking. Our approach is scalable for broader implementation in classrooms and has the potential to be integrated into tools like ChatGPT as an on-boarding experience to encourage learning-oriented use of generative AI.

Recent breakthroughs in artificial intelligence (AI) have brought about increasingly capable systems that demonstrate remarkable abilities in reasoning, language understanding, and problem-solving. These advancements have prompted a renewed examination of AI awareness not as a philosophical question of consciousness, but as a measurable, functional capacity. AI awareness is a double-edged sword: it improves general capabilities, i.e., reasoning, safety, while also raising concerns around misalignment and societal risks, demanding careful oversight as AI capabilities grow. In this review, we explore the emerging landscape of AI awareness, which includes metacognition (the ability to represent and reason about its own cognitive state), self-awareness (recognizing its own identity, knowledge, limitations, inter alia), social awareness (modeling the knowledge, intentions, and behaviors of other agents and social norms), and situational awareness (assessing and responding to the context in which it operates). First, we draw on insights from cognitive science, psychology, and computational theory to trace the theoretical foundations of awareness and examine how the four distinct forms of AI awareness manifest in state-of-the-art AI. Next, we systematically analyze current evaluation methods and empirical findings to better understand these manifestations. Building on this, we explore how AI awareness is closely linked to AI capabilities, demonstrating that more aware AI agents tend to exhibit higher levels of intelligent behaviors. Finally, we discuss the risks associated with AI awareness, including key topics in AI safety, alignment, and broader ethical concerns.

The rise of AI, especially Large Language Models, presents challenges and opportunities to integrate such technology into the classroom. AI has the potential to revolutionize education by helping teaching staff with various tasks, such as personalizing their teaching methods, but it also raises concerns, for example, about the degradation of student-teacher interactions and user privacy. Based on interviews with teaching staff, this paper introduces INSIGHT, a proof of concept to combine various AI tools to assist teaching staff and students in the process of solving exercises. INSIGHT has a modular design that allows it to be integrated into various higher education courses. We analyze students' questions to an LLM by extracting keywords, which we use to dynamically build an FAQ from students' questions and provide new insights for the teaching staff to use for more personalized face-to-face support. Future work could build upon INSIGHT by using the collected data to provide adaptive learning and adjust content based on student progress and learning styles to offer a more interactive and inclusive learning experience.

Lecture slide element detection and retrieval are key problems in slide understanding. Training effective models for these tasks often depends on extensive manual annotation. However, annotating large volumes of lecture slides for supervised training is labor intensive and requires domain expertise. To address this, we propose a large language model (LLM)-guided synthetic lecture slide generation pipeline, SynLec-SlideGen, which produces high-quality, coherent and realistic slides. We also create an evaluation benchmark, namely RealSlide by manually annotating 1,050 real lecture slides. To assess the utility of our synthetic slides, we perform few-shot transfer learning on real data using models pre-trained on them. Experimental results show that few-shot transfer learning with pretraining on synthetic slides significantly improves performance compared to training only on real data. This demonstrates that synthetic data can effectively compensate for limited labeled lecture slides. The code and resources of our work are publicly available on our project website: https://synslidegen.github.io/.

Large Language Models (LLMs) have shown impressive performance on a range of educational tasks, but are still understudied for their potential to solve mathematical problems. In this study, we compare three prominent LLMs, including GPT-4o, DeepSeek-V3, and Gemini-2.0, on three mathematics datasets of varying complexities (GSM8K, MATH500, and MIT Open Courseware datasets). We take a five-dimensional approach based on the Structured Chain-of-Thought (SCoT) framework to assess final answer correctness, step completeness, step validity, intermediate calculation accuracy, and problem comprehension. The results show that GPT-4o is the most stable and consistent in performance across all the datasets, but particularly it performs outstandingly in high-level questions of the MIT Open Courseware dataset. DeepSeek-V3 is competitively strong in well-structured domains such as optimisation, but suffers from fluctuations in accuracy in statistical inference tasks. Gemini-2.0 shows strong linguistic understanding and clarity in well-structured problems but performs poorly in multi-step reasoning and symbolic logic. Our error analysis reveals particular deficits in each model: GPT-4o is at times lacking in sufficient explanation or precision; DeepSeek-V3 leaves out intermediate steps; and Gemini-2.0 is less flexible in mathematical reasoning in higher dimensions.