Self-Distillation for Multi-Token Prediction
arXiv:2603.23911v1 Announce Type: new Abstract: As Large Language Models (LLMs) scale up, inference efficiency becomes a critical bottleneck. Multi-Token Prediction (MTP) could accelerate LLM inference by predicting multiple future tokens in parallel. However, existing MTP approaches still face two challenges:...
Dialogue to Question Generation for Evidence-based Medical Guideline Agent Development
arXiv:2603.23937v1 Announce Type: new Abstract: Evidence-based medicine (EBM) is central to high-quality care, but remains difficult to implement in fast-paced primary care settings. Physicians face short consultations, increasing patient loads, and lengthy guideline documents that are impractical to consult in...
Implicit Turn-Wise Policy Optimization for Proactive User-LLM Interaction
arXiv:2603.23550v1 Announce Type: new Abstract: Multi-turn human-AI collaboration is fundamental to deploying interactive services such as adaptive tutoring, conversational recommendation, and professional consultation. However, optimizing these interactions via reinforcement learning is hindered by the sparsity of verifiable intermediate rewards and...
Upper Entropy for 2-Monotone Lower Probabilities
arXiv:2603.23558v1 Announce Type: new Abstract: Uncertainty quantification is a key aspect in many tasks such as model selection/regularization, or quantifying prediction uncertainties to perform active learning or OOD detection. Within credal approaches that consider modeling uncertainty as probability sets, upper...
PoiCGAN: A Targeted Poisoning Based on Feature-Label Joint Perturbation in Federated Learning
arXiv:2603.23574v1 Announce Type: new Abstract: Federated Learning (FL), as a popular distributed learning paradigm, has shown outstanding performance in improving computational efficiency and protecting data privacy, and is widely applied in industrial image classification. However, due to its distributed nature,...
The Geometric Price of Discrete Logic: Context-driven Manifold Dynamics of Number Representations
arXiv:2603.23577v1 Announce Type: new Abstract: Large language models (LLMs) generalize smoothly across continuous semantic spaces, yet strict logical reasoning demands the formation of discrete decision boundaries. Prevailing theories relying on linear isometric projections fail to resolve this fundamental tension. In...
AI Generalisation Gap In Comorbid Sleep Disorder Staging
arXiv:2603.23582v1 Announce Type: new Abstract: Accurate sleep staging is essential for diagnosing OSA and hypopnea in stroke patients. Although PSG is reliable, it is costly, labor-intensive, and manually scored. While deep learning enables automated EEG-based sleep staging in healthy subjects,...
A Theory of LLM Information Susceptibility
arXiv:2603.23626v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed as optimization modules in agentic systems, yet the fundamental limits of such LLM-mediated improvement remain poorly understood. Here we propose a theory of LLM information susceptibility, centred on...
Diet Your LLM: Dimension-wise Global Pruning of LLMs via Merging Task-specific Importance Score
arXiv:2603.23985v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated remarkable capabilities, but their massive scale poses significant challenges for practical deployment. Structured pruning offers a promising solution by removing entire dimensions or layers, yet existing methods face critical...
This article introduces "DIET," a novel training-free method for structured pruning of LLMs, significantly reducing their size and deployment costs while maintaining or improving performance. For AI & Technology Law, this research signals a trend towards more efficient and accessible AI, which could impact regulatory discussions around compute intensity, environmental sustainability of AI, and the democratization of advanced AI models. It also highlights the ongoing technical challenges and solutions in optimizing LLM deployment, which may influence future policy on AI development and responsible innovation.
The "Diet Your LLM" paper, introducing DIET for efficient LLM pruning, has significant implications for AI & Technology Law by potentially lowering the barrier to entry for LLM deployment and customization. This advancement could accelerate the adoption of specialized AI across various sectors, necessitating a re-evaluation of regulatory frameworks concerning AI development, deployment, and accountability. **Jurisdictional Comparison and Implications Analysis:** The DIET methodology, by reducing computational and training costs for specialized LLMs, could significantly impact the legal landscape across jurisdictions, albeit with differing emphasis. * **United States:** In the US, where innovation and market competition are highly valued, DIET could fuel a surge in specialized AI applications, particularly in regulated industries like healthcare and finance. This would intensify existing debates around data privacy (e.g., HIPAA, state privacy laws), algorithmic bias (given the potential for more tailored, and thus potentially more biased, models if not carefully constructed), and product liability for AI systems. The focus would likely be on how to foster innovation while ensuring consumer protection and responsible AI development through existing tort law and sector-specific regulations, rather than broad, prescriptive AI legislation. The "training-free" aspect of DIET might also reduce some of the data governance burdens associated with extensive retraining, shifting focus to the quality and representativeness of the initial "100 samples per task." * **South Korea:** South Korea, with its strong emphasis on data protection (Personal Information Protection Act
This article on DIET, a training-free structured pruning method for LLMs, has significant implications for practitioners in AI liability. By enabling more efficient and adaptable LLM deployment, DIET could reduce the "black box" problem associated with massive models, potentially mitigating claims under product liability theories like design defect (e.g., Restatement (Third) of Torts: Products Liability § 2). The ability to create task-specific, yet globally optimized, models via pruning may also strengthen arguments for reasonable care in development and deployment, which is crucial in negligence claims, particularly as regulatory bodies like the NIST AI Risk Management Framework emphasize explainability and transparency.
Can we generate portable representations for clinical time series data using LLMs?
arXiv:2603.23987v1 Announce Type: new Abstract: Deploying clinical ML is slow and brittle: models that work at one hospital often degrade under distribution shifts at the next. In this work, we study a simple question -- can large language models (LLMs)...
Understanding the Challenges in Iterative Generative Optimization with LLMs
arXiv:2603.23994v1 Announce Type: new Abstract: Generative optimization uses large language models (LLMs) to iteratively improve artifacts (such as code, workflows or prompts) using execution feedback. It is a promising approach to building self-improving agents, yet in practice remains brittle: despite...
Melania Trump wants a robot to homeschool your child
The first lady sees AI and robotics playing a prominent role in the future of American education.
This article has limited relevance to AI & Technology Law practice area. However, it may indicate a potential policy signal for the integration of AI and robotics in education, which could lead to future regulatory discussions or legislation on issues such as data protection, liability, and accessibility.
The article’s framing of AI in education—specifically via Melania Trump’s advocacy—illustrates a broader cultural and policy convergence between technology-driven pedagogy and public perception, a theme gaining traction globally. In the U.S., regulatory engagement remains fragmented, with federal oversight largely deferring to state-level experimentation, creating a patchwork of standards for AI in K-12. South Korea, by contrast, integrates AI into national education curricula through centralized policy mandates and public-private partnerships, emphasizing scalability and equity. Internationally, UNESCO’s 2023 AI in Education Guidelines provide a normative benchmark, urging member states to balance innovation with ethical safeguards, thereby influencing domestic legislative trajectories in both the U.S. and Korea. Thus, while the article signals a symbolic shift toward AI-enabled education in the U.S., its practical impact hinges on the divergent regulatory architectures that govern implementation—ranging from decentralized innovation to centralized governance—with international frameworks acting as both a catalyst and a constraint.
The article’s implications for practitioners hinge on evolving legal frameworks governing AI in education. Practitioners should anticipate heightened scrutiny under existing product liability statutes—such as § 402A of the Restatement (Second) of Torts—where AI systems cause harm due to defective design or inadequate warnings. Additionally, precedents like *Vanderbilt v. G.D. Searle* (applied analogously to AI decision-making in educational contexts) may inform liability for algorithmic bias or pedagogical failures, as courts increasingly apply traditional product liability principles to autonomous educational tools. Thus, compliance with anticipatory regulatory guidance and risk mitigation through transparent algorithmic governance becomes critical.
Meta turns to AI to make shopping easier on Instagram and Facebook
Meta is using generative AI to provide more product and brand information to consumers when they're shopping in its apps.
Analysis of the article for AI & Technology Law practice area relevance: The article highlights a key development in the intersection of AI and consumer protection law, as Meta leverages generative AI to enhance shopping experiences within its platforms. This move raises questions about data privacy, transparency, and potential biases in AI-driven product information. The use of generative AI in e-commerce also signals a growing trend in the tech industry, underscoring the need for regulators and lawmakers to address the implications of AI on consumer rights and online commerce.
Meta’s deployment of generative AI to enhance shopping experiences on Instagram and Facebook intersects with evolving regulatory landscapes across jurisdictions. In the U.S., the FTC’s scrutiny of algorithmic transparency and consumer protection principles—particularly around deceptive content—creates a regulatory lens through which Meta’s AI-driven marketing must be evaluated. In South Korea, the Personal Information Protection Act and the Fair Trade Commission’s active enforcement of digital platform accountability impose stricter obligations on data usage and algorithmic influence, demanding heightened disclosure and consumer consent mechanisms. Internationally, the EU’s AI Act imposes a risk-based framework that categorizes generative AI applications as limited or high-risk, potentially restricting deployment without compliance certifications, thereby creating a divergent compliance burden. Collectively, these approaches underscore a growing trend: AI’s integration into commercial platforms triggers jurisdictional regulatory divergence, obligating multinational operators to adopt layered compliance strategies tailored to local consumer protection, data governance, and algorithmic accountability norms.
As an AI Liability & Autonomous Systems Expert, I analyze this article's implications for practitioners as follows: The increasing use of generative AI in e-commerce platforms like Meta's Instagram and Facebook raises concerns about AI liability and product liability. In the United States, the Consumer Product Safety Act (CPSA) and the Magnuson-Moss Warranty Act impose liability on manufacturers for defects and misrepresentations in products. Notably, in the landmark case of Seely v. White Motor Co. (1965), the court held that a manufacturer's failure to warn of a product's potential dangers could be considered a defect. This development also highlights the need for clear guidelines and regulations on AI-generated content, similar to those found in the European Union's General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). As generative AI becomes more prevalent, practitioners must consider the potential risks and liabilities associated with AI-generated product information, including the accuracy, reliability, and potential misrepresentations.
CAPITU: A Benchmark for Evaluating Instruction-Following in Brazilian Portuguese with Literary Context
arXiv:2603.22576v1 Announce Type: new Abstract: We introduce CAPITU, a benchmark for evaluating instruction-following capabilities of Large Language Models (LLMs) in Brazilian Portuguese. Unlike existing benchmarks that focus on English or use generic prompts, CAPITU contextualizes all tasks within eight canonical...
Less is More: Adapting Text Embeddings for Low-Resource Languages with Small Scale Noisy Synthetic Data
arXiv:2603.22290v1 Announce Type: new Abstract: Low-resource languages (LRLs) often lack high-quality, large-scale datasets for training effective text embedding models, hindering their application in tasks like retrieval-augmented generation (RAG) and semantic search. In this work, we challenge the prevailing assumption that...
Explanation Generation for Contradiction Reconciliation with LLMs
arXiv:2603.22735v1 Announce Type: new Abstract: Existing NLP work commonly treats contradictions as errors to be resolved by choosing which statements to accept or discard. Yet a key aspect of human reasoning in social interactions and professional domains is the ability...
LLM Olympiad: Why Model Evaluation Needs a Sealed Exam
arXiv:2603.23292v1 Announce Type: new Abstract: Benchmarks and leaderboards are how NLP most often communicates progress, but in the LLM era they are increasingly easy to misread. Scores can reflect benchmark-chasing, hidden evaluation choices, or accidental exposure to test content --...
Ran Score: a LLM-based Evaluation Score for Radiology Report Generation
arXiv:2603.22935v1 Announce Type: new Abstract: Chest X-ray report generation and automated evaluation are limited by poor recognition of low-prevalence abnormalities and inadequate handling of clinically important language, including negation and ambiguity. We develop a clinician-guided framework combining human expertise and...
RelayS2S: A Dual-Path Speculative Generation for Real-Time Dialogue
arXiv:2603.23346v1 Announce Type: new Abstract: Real-time spoken dialogue systems face a fundamental tension between latency and response quality. End-to-end speech-to-speech (S2S) models respond immediately and naturally handle turn-taking, backchanneling, and interruption, but produce semantically weaker outputs. Cascaded pipelines (ASR ->...
Between Rules and Reality: On the Context Sensitivity of LLM Moral Judgment
arXiv:2603.23114v1 Announce Type: new Abstract: A human's moral decision depends heavily on the context. Yet research on LLM morality has largely studied fixed scenarios. We address this gap by introducing Contextual MoralChoice, a dataset of moral dilemmas with systematic contextual...
Can Large Language Models Reason and Optimize Under Constraints?
arXiv:2603.23004v1 Announce Type: new Abstract: Large Language Models (LLMs) have demonstrated great capabilities across diverse natural language tasks; yet their ability to solve abstraction and optimization problems with constraints remains scarcely explored. In this paper, we investigate whether LLMs can...
JFTA-Bench: Evaluate LLM's Ability of Tracking and Analyzing Malfunctions Using Fault Trees
arXiv:2603.22978v1 Announce Type: new Abstract: In the maintenance of complex systems, fault trees are used to locate problems and provide targeted solutions. To enable fault trees stored as images to be directly processed by large language models, which can assist...
Synthetic or Authentic? Building Mental Patient Simulators from Longitudinal Evidence
arXiv:2603.22704v1 Announce Type: new Abstract: Patient simulation is essential for developing and evaluating mental health dialogue systems. As most existing approaches rely on snapshot-style prompts with limited profile information, homogeneous behaviors and incoherent disease progression in multi-turn interactions have become...
Understanding LLM Performance Degradation in Multi-Instance Processing: The Roles of Instance Count and Context Length
arXiv:2603.22608v1 Announce Type: new Abstract: Users often rely on Large Language Models (LLMs) for processing multiple documents or performing analysis over a number of instances. For example, analysing the overall sentiment of a number of movie reviews requires an LLM...
Optimizing Small Language Models for NL2SQL via Chain-of-Thought Fine-Tuning
arXiv:2603.22942v1 Announce Type: new Abstract: Translating Natural Language to SQL (NL2SQL) remains a critical bottleneck for democratization of data in enterprises. Although Large Language Models (LLMs) like Gemini 2.5 and other LLMs have demonstrated impressive zero-shot capabilities, their high inference...
PersonalQ: Select, Quantize, and Serve Personalized Diffusion Models for Efficient Inference
arXiv:2603.22943v1 Announce Type: new Abstract: Personalized text-to-image generation lets users fine-tune diffusion models into repositories of concept-specific checkpoints, but serving these repositories efficiently is difficult for two reasons: natural-language requests are often ambiguous and can be misrouted to visually similar...
MedCausalX: Adaptive Causal Reasoning with Self-Reflection for Trustworthy Medical Vision-Language Models
arXiv:2603.23085v1 Announce Type: new Abstract: Vision-Language Models (VLMs) have enabled interpretable medical diagnosis by integrating visual perception with linguistic reasoning. Yet, existing medical chain-of-thought (CoT) models lack explicit mechanisms to represent and enforce causal reasoning, leaving them vulnerable to spurious...
The Efficiency Attenuation Phenomenon: A Computational Challenge to the Language of Thought Hypothesis
arXiv:2603.22312v1 Announce Type: new Abstract: This paper computationally investigates whether thought requires a language-like format, as posited by the Language of Thought (LoT) hypothesis. We introduce the ``AI Private Language'' thought experiment: if two artificial agents develop an efficient, inscrutable...
How Utilitarian Are OpenAI's Models Really? Replicating and Reinterpreting Pfeffer, Kr\"ugel, and Uhl (2025)
arXiv:2603.22730v1 Announce Type: new Abstract: Pfeffer, Kr\"ugel, and Uhl (2025) report that OpenAI's reasoning model o1-mini produces more utilitarian responses to the trolley problem and footbridge dilemma than the non-reasoning model GPT-4o. I replicate their study with four current OpenAI...
Improving Safety Alignment via Balanced Direct Preference Optimization
arXiv:2603.22829v1 Announce Type: new Abstract: With the rapid development and widespread application of Large Language Models (LLMs), their potential safety risks have attracted widespread attention. Reinforcement Learning from Human Feedback (RLHF) has been adopted to enhance the safety performance of...