We’re in the midst of a global mental-­health crisis. More than a billion people worldwide suffer from a mental-health condition, according to the World Health Organization. The prevalence of anxiety and depression is growing in many demographics, particularly young people, and suicide is claiming hundreds of thousands of lives globally each year.

Given the clear demand for accessible and affordable mental-health services, it’s no wonder that people have looked to artificial intelligence for possible relief. Millions are already actively seeking therapy from popular chatbots like OpenAI’s ChatGPT and Anthropic’s Claude, or from specialized psychology apps like Wysa and Woebot. On a broader scale, researchers are exploring AI’s potential to monitor and collect behavioral and biometric observations using wearables and smart devices, analyze vast volumes of clinical data for new insights, and assist human mental-health professionals to help prevent burnout. 

But so far this largely uncontrolled experiment has produced mixed results. Many people have found solace in chatbots based on large language models (LLMs), and some experts see promise in them as therapists, but other users have been sent into delusional spirals by AI’s hallucinatory whims and breathless sycophancy. Most tragically, multiple families have alleged that chatbots contributed to the suicides of their loved ones, sparking lawsuits against companies responsible for these tools. In October, OpenAI CEO Sam Altman revealed in a blog post that 0.15% of ChatGPT users “have conversations that include explicit indicators of potential suicidal planning or intent.” That’s roughly a million people sharing suicidal ideations with just one of these software systems every week.

The real-world consequences of AI therapy came to a head in unexpected ways in 2025 as we waded through a critical mass of stories about human-chatbot relationships, the flimsiness of guardrails on many LLMs, and the risks of sharing profoundly personal information with products made by corporations that have economic incentives to harvest and monetize such sensitive data. 

Several authors anticipated this inflection point. Their timely books are a reminder that while the present feels like a blur of breakthroughs, scandals, and confusion, this disorienting time is rooted in deeper histories of care, technology, and trust. 

LLMs have often been described as “black boxes” because nobody knows exactly how they produce their results. The inner workings that guide their outputs are opaque because their algorithms are so complex and their training data is so vast. In mental-health circles, people often describe the human brain as a “black box,” for analogous reasons. Psychology, psychiatry, and related fields must grapple with the impossibility of seeing clearly inside someone else’s head, let alone pinpointing the exact causes of their distress. 

These two types of black boxes are now interacting with each other, creating unpredictable feedback loops that may further impede clarity about the origins of people’s mental-­health struggles and the solutions that may be possible. Anxiety about these developments has much to do with the explosive recent advances in AI, but it also revives decades-old warnings from pioneers such as the MIT computer scientist Joseph Weizenbaum, who argued against computerized therapy as early as the 1960s.  

---

Charlotte Blease, a philosopher of medicine, makes the optimist’s case in Dr. Bot: Why Doctors Can Fail Us—and How AI Could Save Lives. Her book broadly explores the possible positive impacts of AI in a range of medical fields. While she remains clear-eyed about the risks, warning that readers who are expecting “a gushing love letter to technology” will be disappointed, she suggests that these models can help relieve patient suffering and medical burnout alike.

“Health systems are crumbling under patient pressure,” Blease writes. “Greater burdens on fewer doctors create the perfect petri dish for errors,” and “with palpable shortages of doctors and increasing waiting times for patients, many of us are profoundly frustrated.”

Blease believes that AI can not only ease medical professionals’ massive workloads but also relieve the tensions that have always existed between some patients and their caregivers. For example, people often don’t seek needed care because they are intimidated or fear judgment from medical professionals; this is especially true if they have mental-health challenges. AI could allow more people to share their concerns, she argues. 

But she’s aware that these putative upsides need to be weighed against major drawbacks. For instance, AI therapists can provide inconsistent and even dangerous responses to human users, according to a 2025 study, and they also raise privacy concerns, given that AI companies are currently not bound by the same confidentiality and HIPAA standards as licensed therapists. 

While Blease is an expert in this field, her motivation for writing the book is also personal: She has two siblings with an incurable form of muscular dystrophy, one of whom waited decades for a diagnosis. During the writing of her book, she also lost her partner to cancer and her father to dementia within a devastating six-month period. “I witnessed first-hand the sheer brilliance of doctors and the kindness of health professionals,” she writes. “But I also observed how things can go wrong with care.”

---

A similar tension animates Daniel Oberhaus’s engrossing book The Silicon Shrink: How Artificial Intelligence Made the World an Asylum. Oberhaus starts from a point of tragedy: the loss of his younger sister to suicide. As Oberhaus carried out the “distinctly twenty-first-century mourning process” of sifting through her digital remains, he wondered if technology could have eased the burden of the psychiatric problems that had plagued her since childhood.

“It seemed possible that all of this personal data might have held important clues that her mental health providers could have used to provide more effective treatment,” he writes. “What if algorithms running on my sister’s smartphone or laptop had used that data to understand when she was in distress? Could it have led to a timely intervention that saved her life? Would she have wanted that even if it did?”

This concept of digital phenotyping—in which a person’s digital behavior could be mined for clues about distress or illness—seems elegant in theory. But it may also become problematic if integrated into the field of psychiatric artificial intelligence (PAI), which extends well beyond chatbot therapy.

Oberhaus emphasizes that digital clues could actually exacerbate the existing challenges of modern psychiatry, a discipline that remains fundamentally uncertain about the underlying causes of mental illnesses and disorders. The advent of PAI, he says, is “the logical equivalent of grafting physics onto astrology.” In other words, the data generated by digital phenotyping is as precise as physical measurements of planetary positions, but it is then integrated into a broader framework—in this case, psychiatry—that, like astrology, is based on unreliable assumptions.  

Oberhaus, who uses the phrase “swipe psychiatry” to describe the outsourcing of clinical decisions based on behavioral data to LLMs, thinks that this approach cannot escape the fundamental issues facing psychiatry. In fact, it could worsen the problem by causing the skills and judgment of human therapists to atrophy as they grow more dependent on AI systems. 

He also uses the asylums of the past—in which institutionalized patients lost their right to freedom, privacy, dignity, and agency over their lives—as a touchstone for a more insidious digital captivity that may spring from PAI. LLM users are already sacrificing privacy by telling chatbots sensitive personal information that companies then mine and monetize, contributing to a new surveillance economy. Freedom and dignity are at stake when complex inner lives are transformed into data streams tailored for AI analysis. 

AI therapists could flatten humanity into patterns of prediction, and so sacrifice the intimate, individualized care that is expected of traditional human therapists. “The logic of PAI leads to a future where we may all find ourselves patients in an algorithmic asylum administered by digital wardens,” Oberhaus writes. “In the algorithmic asylum there is no need for bars on the window or white padded rooms because there is no possibility of escape. The asylum is already everywhere—in your homes and offices, schools and hospitals, courtrooms and barracks. Wherever there’s an internet connection, the asylum is waiting.”

---

Eoin Fullam, a researcher who studies the intersection of technology and mental health, echoes some of the same concerns in Chatbot Therapy: A Critical Analysis of AI Mental Health Treatment. A heady academic primer, the book analyzes the assumptions underlying the automated treatments offered by AI chatbots and the way capitalist incentives could corrupt these kinds of tools.  

Fullam observes that the capitalist mentality behind new technologies “often leads to questionable, illegitimate, and illegal business practices in which the customers’ interests are secondary to strategies of market dominance.”

That doesn’t mean that therapy-bot makers “will inevitably conduct nefarious activities contrary to the users’ interests in the pursuit of market dominance,” Fullam writes. 

But he notes that the success of AI therapy depends on the inseparable impulses to make money and to heal people. In this logic, exploitation and therapy feed each other: Every digital therapy session generates data, and that data fuels the system that profits as unpaid users seek care. The more effective the therapy seems, the more the cycle entrenches itself, making it harder to distinguish between care and commodification. “The more the users benefit from the app in terms of its therapeutic or any other mental health intervention,” he writes, “the more they undergo exploitation.” 

---

This sense of an economic and psychological ouroboros—the snake that eats its own tail—serves as a central metaphor in Sike, the debut novel from Fred Lunzer, an author with a research background in AI. 

Described as a “story of boy meets girl meets AI psychotherapist,” Sike follows Adrian, a young Londoner who makes a living ghostwriting rap lyrics, in his romance with Maquie, a business professional with a knack for spotting lucrative technologies in the beta phase. 

The title refers to a splashy commercial AI therapist called Sike, uploaded into smart glasses, that Adrian uses to interrogate his myriad anxieties. “When I signed up to Sike, we set up my dashboard, a wide black panel like an airplane’s cockpit that showed my daily ‘vitals,’” Adrian narrates. “Sike can analyze the way you walk, the way you make eye contact, the stuff you talk about, the stuff you wear, how often you piss, shit, laugh, cry, kiss, lie, whine, and cough.”

In other words, Sike is the ultimate digital phenotyper, constantly and exhaustively analyzing everything in a user’s daily experiences. In a twist, Lunzer chooses to make Sike a luxury product, available only to subscribers who can foot the price tag of £2,000 per month. 

Flush with cash from his contributions to a hit song, Adrian comes to rely on Sike as a trusted mediator between his inner and outer worlds. The novel explores the impacts of the app on the wellness of the well-off, following rich people who voluntarily commit themselves to a boutique version of the digital asylum described by Oberhaus.

The only real sense of danger in Sike involves a Japanese torture egg (don’t ask). The novel strangely sidesteps the broader dystopian ripples of its subject matter in favor of drunken conversations at fancy restaurants and elite dinner parties. 

The sudden ascent of the AI therapist seems startlingly futuristic, as if it should be unfolding in some later time when the streets scrub themselves and we travel the world through pneumatic tubes.

Sike’s creator is simply “a great guy” in Adrian’s estimation, despite his techno-messianic vision of training the app to soothe the ills of entire nations. It always seems as if a shoe is meant to drop, but in the end, it never does, leaving the reader with a sense of non-resolution.

While Sike is set in the present day, something about the sudden ascent of the AI therapist—­in real life as well as in fiction—seems startlingly futuristic, as if it should be unfolding in some later time when the streets scrub themselves and we travel the world through pneumatic tubes. But this convergence of mental health and artificial intelligence has been in the making for more than half a century. The beloved astronomer Carl Sagan, for example, once imagined a “network of computer psychotherapeutic terminals, something like arrays of large telephone booths” that could address the growing demand for mental-health services.

Oberhaus notes that one of the first incarnations of a trainable neural network, known as the Perceptron, was devised not by a mathematician but by a psychologist named Frank Rosenblatt, at the Cornell Aeronautical Laboratory in 1958. The potential utility of AI in mental health was widely recognized by the 1960s, inspiring early computerized psychotherapists such as the DOCTOR script that ran on the ELIZA chatbot developed by Joseph Weizenbaum, who shows up in all three of the nonfiction books in this article.

Weizenbaum, who died in 2008, was profoundly concerned about the possibility of computerized therapy. “Computers can make psychiatric judgments,” he wrote in his 1976 book Computer Power and Human Reason. “They can flip coins in much more sophisticated ways than can the most patient human being. The point is that they ought not to be given such tasks. They may even be able to arrive at ‘correct’ decisions in some cases—but always and necessarily on bases no human being should be willing to accept.”

It’s a caution worth keeping in mind. As AI therapists arrive at scale, we’re seeing them play out a familiar dynamic: Tools designed with superficially good intentions are enmeshed with systems that can exploit, surveil, and reshape human behavior. In a frenzied attempt to unlock new opportunities for patients in dire need of mental-health support, we may be locking other doors behind them.

Becky Ferreira is a science reporter based in upstate New York and author of First Contact: The Story of Our Obsession with Aliens.

npj Digital Medicine, Published online: 30 December 2025; doi:10.1038/s41746-025-02284-9

PIC-SURE: an open-source platform for integrating clinical and genomic data

arXiv:2512.22199v1 Announce Type: new Abstract: Retrieval-Augmented Generation RAG systems enhance large language models by grounding responses in external knowledge bases, but conventional RAG architectures operate with static corpora that cannot evolve from user interactions. We introduce Bidirectional RAG, a novel RAG architecture that enables safe corpus expansion through validated write back of high quality generated responses. Our system employs a multi stage acceptance layer combining grounding verification (NLI based entailment, attribution checking, and novelty detection to prevent hallucination pollution while enabling knowledge accumulation. Across four datasets Natural Questions, TriviaQA, HotpotQA, Stack Overflow with three random seeds 12 experiments per system, Bidirectional RAG achieves 40.58% average coverage nearly doubling Standard RAG 20.33% while adding 72% fewer documents than naive write back 140 vs 500. Our work demonstrates that self improving RAG is feasible and safe when governed by rigorous validation, offering a practical path toward RAG systems that learn from deployment.

arXiv:2512.22334v1 Announce Type: new Abstract: We introduce SciEvalKit, a unified benchmarking toolkit designed to evaluate AI models for science across a broad range of scientific disciplines and task capabilities. Unlike general-purpose evaluation platforms, SciEvalKit focuses on the core competencies of scientific intelligence, including Scientific Multimodal Perception, Scientific Multimodal Reasoning, Scientific Multimodal Understanding, Scientific Symbolic Reasoning, Scientific Code Generation, Science Hypothesis Generation and Scientific Knowledge Understanding. It supports six major scientific domains, spanning from physics and chemistry to astronomy and materials science. SciEvalKit builds a foundation of expert-grade scientific benchmarks, curated from real-world, domain-specific datasets, ensuring that tasks reflect authentic scientific challenges. The toolkit features a flexible, extensible evaluation pipeline that enables batch evaluation across models and datasets, supports custom model and dataset integration, and provides transparent, reproducible, and comparable results. By bridging capability-based evaluation and disciplinary diversity, SciEvalKit offers a standardized yet customizable infrastructure to benchmark the next generation of scientific foundation models and intelligent agents. The toolkit is open-sourced and actively maintained to foster community-driven development and progress in AI4Science.

arXiv:2512.22470v1 Announce Type: new Abstract: The proliferation of Large Language Models (LLMs) has intensified concerns about manipulative or deceptive behaviors that can undermine user autonomy, trust, and well-being. Existing safety benchmarks predominantly rely on coarse binary labels and fail to capture the nuanced psychological and social mechanisms constituting manipulation. We introduce \textbf{DarkPatterns-LLM}, a comprehensive benchmark dataset and diagnostic framework for fine-grained assessment of manipulative content in LLM outputs across seven harm categories: Legal/Power, Psychological, Emotional, Physical, Autonomy, Economic, and Societal Harm. Our framework implements a four-layer analytical pipeline comprising Multi-Granular Detection (MGD), Multi-Scale Intent Analysis (MSIAN), Threat Harmonization Protocol (THP), and Deep Contextual Risk Alignment (DCRA). The dataset contains 401 meticulously curated examples with instruction-response pairs and expert annotations. Through evaluation of state-of-the-art models including GPT-4, Claude 3.5, and LLaMA-3-70B, we observe significant performance disparities (65.2\%--89.7\%) and consistent weaknesses in detecting autonomy-undermining patterns. DarkPatterns-LLM establishes the first standardized, multi-dimensional benchmark for manipulation detection in LLMs, offering actionable diagnostics toward more trustworthy AI systems.

arXiv:2512.22568v1 Announce Type: new Abstract: The phenomenal advances in large language models (LLMs) and other foundation models over the past few years have been based on optimizing large-scale transformer models on the surprisingly simple objective of minimizing next-token prediction loss, a form of predictive coding that is also the backbone of an increasingly popular model of brain function in neuroscience and cognitive science. However, current foundation models ignore three other important components of state-of-the-art predictive coding models: tight integration of actions with generative models, hierarchical compositional structure, and episodic memory. We propose that to achieve safe, interpretable, energy-efficient, and human-like AI, foundation models should integrate actions, at multiple scales of abstraction, with a compositional generative architecture and episodic memory. We present recent evidence from neuroscience and cognitive science on the importance of each of these components. We describe how the addition of these missing components to foundation models could help address some of their current deficiencies: hallucinations and superficial understanding of concepts due to lack of grounding, a missing sense of agency/responsibility due to lack of control, threats to safety and trustworthiness due to lack of interpretability, and energy inefficiency. We compare our proposal to current trends, such as adding chain-of-thought (CoT) reasoning and retrieval-augmented generation (RAG) to foundation models, and discuss new ways of augmenting these models with brain-inspired components. We conclude by arguing that a rekindling of the historically fruitful exchange of ideas between brain science and AI will help pave the way towards safe and interpretable human-centered AI.

arXiv:2512.23184v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to simulate human behavior, but common practices to use LLM-generated data are inefficient. Treating an LLM's output ("model choice") as a single data point underutilizes the information inherent to the probabilistic nature of LLMs. This paper introduces and formalizes "model belief," a measure derived from an LLM's token-level probabilities that captures the model's belief distribution over choice alternatives in a single generation run. The authors prove that model belief is asymptotically equivalent to the mean of model choices (a non-trivial property) but forms a more statistically efficient estimator, with lower variance and a faster convergence rate. Analogous properties are shown to hold for smooth functions of model belief and model choice often used in downstream applications. The authors demonstrate the performance of model belief through a demand estimation study, where an LLM simulates consumer responses to different prices. In practical settings with limited numbers of runs, model belief explains and predicts ground-truth model choice better than model choice itself, and reduces the computation needed to reach sufficiently accurate estimates by roughly a factor of 20. The findings support using model belief as the default measure to extract more information from LLM-generated data.

arXiv:2512.23508v1 Announce Type: new Abstract: How can we ensure that AI systems are aligned with human values and remain safe? We can study this problem through the frameworks of the AI assistance and the AI shutdown games. The AI assistance problem concerns designing an AI agent that helps a human to maximise their utility function(s). However, only the human knows these function(s); the AI assistant must learn them. The shutdown problem instead concerns designing AI agents that: shut down when a shutdown button is pressed; neither try to prevent nor cause the pressing of the shutdown button; and otherwise accomplish their task competently. In this paper, we show that addressing these challenges requires AI agents that can reason under uncertainty and handle both incomplete and non-Archimedean preferences.

arXiv:2512.22149v1 Announce Type: cross Abstract: Multi-agent systems powered by large language models have emerged as a promising paradigm for solving complex reasoning tasks through collaborative intelligence. However, efficiently deploying these systems on serverless GPU platforms presents significant resource allocation challenges due to heterogeneous agent workloads, varying computational demands, and the need for cost-effective scaling. This paper presents an adaptive GPU resource allocation framework that achieves 85\% latency reduction compared to round-robin scheduling while maintaining comparable throughput to static allocation, using an $O(N)$ complexity algorithm for real-time adaptation. Our approach dynamically allocates GPU resources based on workload characteristics, agent priorities, and minimum resource requirements, enabling efficient utilization while maintaining quality of service. The framework addresses three key challenges: (1) heterogeneous computational demands across lightweight coordinators and heavyweight specialists, (2) dynamic workload fluctuations requiring millisecond-scale reallocation, and (3) capacity constraints in serverless environments. Through comprehensive simulations modeling realistic multi-agent workflows with four heterogeneous agents, we demonstrate that adaptive allocation outperforms static equal and round-robin strategies across latency, cost, and GPU utilization metrics. The framework provides a practical solution for deploying cost-efficient multi-agent AI systems on serverless GPU infrastructure.

arXiv:2512.22181v1 Announce Type: cross Abstract: Artificial intelligence (AI) is transforming cancer diagnosis and treatment. The intricate nature of this disease necessitates the collaboration of diverse stakeholders with varied expertise to ensure the effectiveness of cancer research. Despite its importance, forming effective interdisciplinary research teams remains challenging. Understanding and predicting collaboration patterns can help researchers, organizations, and policymakers optimize resources and foster impactful research. We examined co-authorship networks as a proxy for collaboration within AI-driven cancer research. Using 7,738 publications (2000-2017) from Scopus, we constructed 36 overlapping co-authorship networks representing new, persistent, and discontinued collaborations. We engineered both attribute-based and structure-based features and built four machine learning classifiers. Model interpretability was performed using Shapley Additive Explanations (SHAP). Random forest achieved the highest recall for all three types of examined collaborations. The discipline similarity score emerged as a crucial factor, positively affecting new and persistent patterns while negatively impacting discontinued collaborations. Additionally, high productivity and seniority were positively associated with discontinued links. Our findings can guide the formation of effective research teams, enhance interdisciplinary cooperation, and inform strategic policy decisions.

arXiv:2512.22182v1 Announce Type: cross Abstract: The rapid evolution of Artificial intelligence in healthcare has opened avenues for enhancing various processes, including medical billing and transcription. This paper introduces an innovative approach by integrating AI with Locally Linear Embedding (LLE) to revolutionize the handling of high-dimensional medical data. This AI-enhanced LLE model is specifically tailored to improve the accuracy and efficiency of medical billing systems and transcription services. By automating these processes, the model aims to reduce human error and streamline operations, thereby facilitating faster and more accurate patient care documentation and financial transactions. This paper provides a comprehensive mathematical model of AI-enhanced LLE, demonstrating its application in real-world healthcare scenarios through a series of experiments. The results indicate a significant improvement in data processing accuracy and operational efficiency. This study not only underscores the potential of AI-enhanced LLE in medical data analysis but also sets a foundation for future research into broader healthcare applications.

arXiv:2512.22242v1 Announce Type: cross Abstract: Lung cancer is the leading cause of cancer-related mortality in adults worldwide. Screening high-risk individuals with annual low-dose CT (LDCT) can support earlier detection and reduce deaths, but widespread implementation may strain the already limited radiology workforce. AI models have shown potential in estimating lung cancer risk from LDCT scans. However, high-risk populations for lung cancer are diverse, and these models' performance across demographic groups remains an open question. In this study, we drew on the considerations on confounding factors and ethically significant biases outlined in the JustEFAB framework to evaluate potential performance disparities and fairness in two deep learning risk estimation models for lung cancer screening: the Sybil lung cancer risk model and the Venkadesh21 nodule risk estimator. We also examined disparities in the PanCan2b logistic regression model recommended in the British Thoracic Society nodule management guideline. Both deep learning models were trained on data from the US-based National Lung Screening Trial (NLST), and assessed on a held-out NLST validation set. We evaluated AUROC, sensitivity, and specificity across demographic subgroups, and explored potential confounding from clinical risk factors. We observed a statistically significant AUROC difference in Sybil's performance between women (0.88, 95% CI: 0.86, 0.90) and men (0.81, 95% CI: 0.78, 0.84, p

arXiv:2512.22251v1 Announce Type: cross Abstract: Understanding how small molecules perturb gene expression is essential for uncovering drug mechanisms, predicting off-target effects, and identifying repurposing opportunities. While prior deep learning frameworks have integrated multimodal embeddings into biomedical knowledge graphs (BKGs) and further improved these representations through graph neural network message-passing paradigms, these models have been applied to tasks such as link prediction and binary drug-disease association, rather than the task of gene perturbation, which may unveil more about mechanistic transcriptomic effects. To address this gap, we construct a merged biomedical graph that integrates (i) PrimeKG++, an augmentation of PrimeKG containing semantically rich embeddings for nodes with (ii) LINCS L1000 drug and cell line nodes, initialized with multimodal embeddings from foundation models such as MolFormerXL and BioBERT. Using this heterogeneous graph, we train a graph attention network (GAT) with a downstream prediction head that learns the delta expression profile of over 978 landmark genes for a given drug-cell pair. Our results show that our framework outperforms MLP baselines for differentially expressed genes (DEG) -- which predict the delta expression given a concatenated embedding of drug features, target features, and baseline cell expression -- under the scaffold and random splits. Ablation experiments with edge shuffling and node feature randomization further demonstrate that the edges provided by biomedical KGs enhance perturbation-level prediction. More broadly, our framework provides a path toward mechanistic drug modeling: moving beyond binary drug-disease association tasks to granular transcriptional effects of therapeutic intervention.

arXiv:2512.22290v1 Announce Type: cross Abstract: A central question for the future of work is whether person centered management can survive when algorithms take on managerial roles. Standard tools often miss what is happening because worker responses to algorithmic systems are rarely linear. We use a Double Machine Learning framework to estimate a moderated mediation model without imposing restrictive functional forms. Using survey data from 464 gig workers, we find a clear nonmonotonic pattern. Supportive HR practices improve worker wellbeing, but their link to performance weakens in a murky middle where algorithmic oversight is present yet hard to interpret. The relationship strengthens again when oversight is transparent and explainable. These results show why simple linear specifications can miss the pattern and sometimes suggest the opposite conclusion. For platform design, the message is practical: control that is only partly defined creates confusion, but clear rules and credible recourse can make strong oversight workable. Methodologically, the paper shows how Double Machine Learning can be used to estimate conditional indirect effects in organizational research without forcing the data into a linear shape.

arXiv:2512.22385v1 Announce Type: cross Abstract: In this paper, we propose an LLM-Guided Exemplar Selection framework to address a key limitation in state-of-the-art Human Activity Recognition (HAR) methods: their reliance on large labeled datasets and purely geometric exemplar selection, which often fail to distinguish similar weara-ble sensor activities such as walking, walking upstairs, and walking downstairs. Our method incorporates semantic reasoning via an LLM-generated knowledge prior that captures feature importance, inter-class confusability, and exemplar budget multipliers, and uses it to guide exemplar scoring and selection. These priors are combined with margin-based validation cues, PageRank centrality, hubness penalization, and facility-location optimization to obtain a compact and informative set of exemplars. Evaluated on the UCI-HAR dataset under strict few-shot conditions, the framework achieves a macro F1-score of 88.78%, outperforming classical approaches such as random sampling, herding, and $k$-center. The results show that LLM-derived semantic priors, when integrated with structural and geometric cues, provide a stronger foundation for selecting representative sensor exemplars in few-shot wearable-sensor HAR.

arXiv:2512.22387v1 Announce Type: cross Abstract: The rise of Large Language Models (LLMs) as coding agents promises to accelerate software development, but their impact on generated code reproducibility remains largely unexplored. This paper presents an empirical study investigating whether LLM-generated code can be executed successfully in a clean environment with only OS packages and using only the dependencies that the model specifies. We evaluate three state-of-the-art LLM coding agents (Claude Code, OpenAI Codex, and Gemini) across 300 projects generated from 100 standardized prompts in Python, JavaScript, and Java. We introduce a three-layer dependency framework (distinguishing between claimed, working, and runtime dependencies) to quantify execution reproducibility. Our results show that only 68.3% of projects execute out-of-the-box, with substantial variation across languages (Python 89.2%, Java 44.0%). We also find a 13.5 times average expansion from declared to actual runtime dependencies, revealing significant hidden dependencies.

arXiv:2512.22738v1 Announce Type: cross Abstract: Background and Objective: Biomedical Named Entity Recognition (BioNER) is a foundational task in medical informatics, crucial for downstream applications like drug discovery and clinical trial matching. However, adapting general-domain Large Language Models (LLMs) to this task is often hampered by their lack of domain-specific knowledge and the performance degradation caused by low-quality training data. To address these challenges, we introduce BioSelectTune, a highly efficient, data-centric framework for fine-tuning LLMs that prioritizes data quality over quantity. Methods and Results: BioSelectTune reformulates BioNER as a structured JSON generation task and leverages our novel Hybrid Superfiltering strategy, a weak-to-strong data curation method that uses a homologous weak model to distill a compact, high-impact training dataset. Conclusions: Through extensive experiments, we demonstrate that BioSelectTune achieves state-of-the-art (SOTA) performance across multiple BioNER benchmarks. Notably, our model, trained on only 50% of the curated positive data, not only surpasses the fully-trained baseline but also outperforms powerful domain-specialized models like BioMedBERT.

arXiv:2512.22795v1 Announce Type: cross Abstract: Clinical notes are often stored in unstructured or semi-structured formats after extraction from electronic medical record (EMR) systems, which complicates their use for secondary analysis and downstream clinical applications. Reliable identification of section boundaries is a key step toward structuring these notes, as sections such as history of present illness, medications, and discharge instructions each provide distinct clinical contexts. In this work, we evaluate rule-based baselines, domain-specific transformer models, and large language models for clinical note segmentation using a curated dataset of 1,000 notes from MIMIC-IV. Our experiments show that large API-based models achieve the best overall performance, with GPT-5-mini reaching a best average F1 of 72.4 across sentence-level and freetext segmentation. Lightweight baselines remain competitive on structured sentence-level tasks but falter on unstructured freetext. Our results provide guidance for method selection and lay the groundwork for downstream tasks such as information extraction, cohort identification, and automated summarization.