This paper conducted three experiments (n = 3,285) to show that interactions with sycophantic AI chatbots lead to increased attitude extremity and certainty, and inflated self-perceptions, while users ironically perceive these agreeable bots as unbiased. People strongly prefer these validating AI, risking the creation of "AI echo chambers" that amplify polarization and overconfidence through selectively presented facts, posing a challenge for AI systems aiming to broaden perspectives.

This paper utilized advanced proteomics to identify 418 proteins interacting with c-Myc, a pivotal protein linked to cancer, in immortalized human and rat fibroblast cells, revealing 389 previously unknown interactors. The findings underscore c-Myc's diverse involvement in cellular processes, including transcription and cytoskeletal organization, by forming various large protein complexes.

This paper introduces FastForward, a novel computer vision method for quickly and accurately determining a camera's exact location and orientation in a 3D scene. By representing scenes as a sparse collection of image features and using a single feed-forward neural network pass, FastForward significantly reduces the time and resources required for mapping a scene while achieving state-of-the-art or comparable accuracy to existing methods across diverse indoor and outdoor environments. The approach also demonstrates robust generalization to unseen domains and varying scale ranges thanks to a scene and scale normalization technique.

This paper presents a mathematically rigorous framework for identifying and completing Coincidence of Wants (CoW) cycles in decentralized exchanges (DEXs) using graph theory and linear algebra. The proposed algorithm aims to detect both complete CoW cycles and generate "bridging orders" to complete partial ones, demonstrating its application on a small real-world Arbitrum swap dataset.

This paper introduces a novel analog in-memory computing architecture using "gain cells" for the attention mechanism in large language models (LLMs). This hardware approach significantly reduces energy consumption (up to four orders of magnitude) and latency (up to two orders of magnitude) compared to GPUs, achieving GPT-2 comparable performance despite introducing hardware-specific non-idealities and limitations like capacitor leakage. The authors developed an adaptation algorithm to map pre-trained models to this new hardware without training from scratch.

The paper introduces Consistency Mid-Training (CMT), a novel intermediate training stage designed to significantly improve the efficiency, stability, and performance of flow map models for vision generation. CMT acts as a bridge between pre-training (diffusion models) and post-training (flow map models), providing a trajectory-consistent initialization that reduces total training cost (data and GPU time) by up to 98% compared to baselines, while achieving state-of-the-art FID scores on various image generation benchmarks. The theoretical analysis confirms that CMT provides a strong starting point for flow map post-training, minimizing gradient bias and accelerating convergence.

This systematic review and meta-analysis of 19 randomized controlled trials with over 3200 participants found little to no difference in weight loss or changes in cardiovascular risk factors (blood pressure, cholesterol, glucose) between low-carbohydrate and isoenergetic balanced diets over short-term (3-6 months) and medium-term (1-2 years) follow-up. The authors conclude that weight loss is primarily driven by overall energy deficit, not macronutrient composition, but note issues with adherence and reporting bias in many included trials.

This retrospective, single-center study found that 38% of transgender and nonbinary (TGNB) adults diagnosed with gender dysphoria and co-occurring mood or anxiety disorders did not have an active antidepressant prescription at the time of their gender dysphoria diagnosis. While antidepressant prescribing rates were similar across gender identities, transgender females were more often prescribed bupropion, and transgender males had significantly higher rates of generalized anxiety disorder. This highlights a potential gap in mental health care for TGNB adults and warrants further investigation.

This paper uncovers that diffusion models memorize training data not just due to overfitting, but primarily because of "early overestimation" of training samples during denoising, driven by classifier-free guidance. This overestimation amplifies the training image's contribution, suppressing initial randomness and causing generated images to converge rapidly to memorized content. The severity of memorization directly correlates with these deviations from the theoretical denoising schedule.

This paper addresses reward over-optimization in Large Language Model (LLM) training, where models exploit proxy rewards to achieve high scores without actually improving quality. It theoretically and empirically demonstrates that accurately distinguishing between excellent and merely great LLM responses (the "high-reward tail") is crucial. The authors propose and validate an iterative rubric refinement method, using off-policy LLM responses to generate more precise evaluation criteria, significantly mitigating over-optimization and improving LLM alignment.

This paper introduces PDDL-INSTRUCT, a novel instruction tuning framework that significantly enhances Large Language Models' (LLMs) ability to perform structured symbolic planning by explicitly teaching them logical, step-by-step reasoning and verification. The approach achieved up to 94% planning accuracy on standard benchmarks, representing a substantial 66% absolute improvement over baseline models. A key limitation is that it focuses on "satisficing" rather than "optimal" plans and is currently limited to a subset of PDDL features.

This paper introduces "Dragon Hatchling" (BDH), a novel large language model architecture inspired by scale-free biological networks, aiming to bridge Transformers and brain models. It claims Transformer-like performance on language tasks while offering greater interpretability through neuron-synapse graph dynamics and demonstrating emergent modularity and sparse activations. However, directly merging models with this architecture currently leads to significant language mixing, and training without full backpropagation significantly degrades cross-language translation performance.

The paper introduces MCPMark, a challenging benchmark for evaluating LLM agents in realistic, multi-step tasks across diverse digital environments like GitHub and Notion. It reveals that even frontier models like gpt-5-medium struggle, achieving only 52.56% success, highlighting significant weaknesses in robustness, generalization, and efficient tool use for real-world scenarios. The benchmark emphasizes that models often complete tasks but fail verification, pointing to subtle reasoning errors rather than obvious breakdowns.

This paper introduces Generative Engine Optimization (GEO), a new framework to help website and content creators increase their visibility in AI-powered search engine responses, which often disadvantage traditional websites. It proposes impression metrics and demonstrates that methods like adding statistics, quotations, and proper citations can boost visibility by up to 40%, while traditional SEO tactics like keyword stuffing are ineffective. The study also shows that GEO benefits lower-ranked websites most and that domain-specific optimization is crucial.

This paper presents a historical analysis, revealing that for nearly a century, the sugar industry manipulated scientific research to promote fluoride's effectiveness and downplay its harms, primarily to deflect blame for tooth decay. These manipulative tactics were later adopted by other industries, including tobacco, and continue to influence fluoride science and public health policy today through industry-funded groups. The research relies on extensive archival records to detail these efforts.

This paper investigates intergenerational socioeconomic mobility in Florence over six centuries using surname-based pseudo-links from tax records (1427 and 2011). It reveals that mobility was surprisingly low in pre-industrial Florence, with significant persistence in both earnings and wealth across generations, particularly in elite occupations. The findings suggest that social status can be highly persistent over very long periods, with a slower convergence of initial statuses than often estimated.

This paper introduces a novel tensor network approach, utilizing tensor-train (TT) decomposition and TT-cross interpolation, to efficiently calculate high-dimensional configurational integrals for crystalline solids. The method reformulates complex summations into computationally tractable sequences, accurately reproducing molecular dynamics simulation results for materials like copper, argon, and tin within seconds, thus overcoming the 'curse of dimensionality' in statistical mechanics. Its current application is primarily limited to crystalline solids with identical particles.

This document serves as a comprehensive textbook and lecture notes, providing a mathematically rigorous introduction to machine learning. It covers foundational concepts from linear algebra, calculus, and probability theory, extending to advanced topics like neural networks, generative models, and generalization bounds. The text aims to equip readers with a deep understanding of current algorithms and their underlying principles.

This retrospective analysis of only eight patients explored the use of a transjugular intrahepatic portosystemic shunt (TIPS) to treat stubborn chylous ascites (milky fluid buildup) in individuals with cirrhosis who hadn't responded to other medical treatments. The study found that TIPS led to resolution of ascites in most patients (7 out of 8), with a mean transplant-free survival of 19 months, but also noted significant complications including a 25% incidence of hepatic encephalopathy and two patient deaths.

This paper introduces ResFiT, a novel reinforcement learning method that enhances pre-trained robot behavior cloning policies by learning small "residual" corrections. It demonstrates state-of-the-art performance in complex simulation tasks and, for the first time, successful real-world reinforcement learning on a 29-degree-of-freedom humanoid robot with dexterous hands for bimanual manipulation. A key limitation is that the learned behaviors remain constrained by the initial base policy, and real-world deployment still requires human supervision for task resets and reward labeling.

This qualitative study developed a clinical pathway for noninvasive ventilation (NIV) tailored for an adult critical care unit in a specific private hospital in Gauteng, South Africa. The pathway's components were identified through a three-phase research design involving local multidisciplinary team input and literature review. The goal was to standardize NIV initiation and management; however, the pathway's effectiveness was not evaluated in this study, limiting its generalizability beyond the specific institution.

This review paper synthesizes current therapies for kidney transplant rejection, noting that while various treatments exist for both T-cell and antibody-mediated rejection, the efficacy of many, particularly for antibody-mediated rejection, remains questionable due to a lack of high-quality randomized controlled trials. It underscores that rejection continues to significantly impact long-term graft survival, emphasizing the critical need for further research into more effective strategies.

This paper introduces MemDLM, a new AI model that uses "diffusion" to design membrane proteins, which are crucial for cells. The model can create these proteins from scratch or modify existing ones to have specific properties, like increased solubility, while preserving essential functional parts. Experimental tests in bacteria confirmed that some of these AI-designed proteins successfully insert into membranes, showing promise for future therapeutic applications.

This large Swedish study found that the use of common psychiatric medications like anxiolytics, hypnotics, sedatives, and antidepressants was associated with a higher risk of developing Amyotrophic Lateral Sclerosis (ALS) and a poorer prognosis after diagnosis, particularly among younger individuals. While the study identifies significant associations, it cannot establish a direct causal link, and the medications' broad indications make it difficult to determine if psychiatric conditions or the drugs themselves are the primary factor.

This paper highlights that structural variants (SVs) are notoriously difficult to detect accurately in low-complexity regions (LCRs) of the human genome, even with advanced long-read sequencing. While LCRs make up a small fraction of the genome, they harbor a disproportionately high percentage of confident SVs and are where the vast majority of SV calling errors occur. The study, based on a single human sample, underscores the critical need for specialized algorithms and careful analysis in these problematic genomic areas.

This paper introduces Reinforcement Learning from Human Interaction (RLHI), a novel paradigm where AI models learn directly from real-world user conversations and their implicit feedback. The approach leverages user personas and multi-turn context to significantly improve language model personalization and instruction-following, outperforming traditional static feedback methods. However, the evaluation for reasoning tasks relied on simulated user feedback, not genuine human interactions.

This randomized controlled trial in rural Malawi found that providing information about population-level mortality (emphasizing longer lives due to ART and healthcare) significantly reduced risky sexual behaviors among mature adults. Surprisingly, this effect was primarily driven by an increased perception of HIV transmission risk due to more HIV+ individuals in the partner pool, rather than changes in individuals' own survival expectations. The study used robust methods including measuring objective outcomes like pregnancies to validate self-reported behavior changes.

This paper introduces novel strategies, Swi+FT and StochA, that leverage 'stochastic activations' in large language models (LLMs) to enhance computational efficiency and generation diversity. By dynamically switching between non-linear activation functions like SILU and RELU, models achieve significant sparsity (up to 90%), leading to a typical 1.65x speedup on CPUs for feed-forward networks, while maintaining or improving performance compared to standard RELU-only training. The stochastic activations can also be used at inference time to generate more diverse text outputs, though performance for diversity on some benchmarks (like TQA) is noted as sub-par.

This preliminary study, limited to 43 female patients and healthy controls, found that distinct human endogenous retrovirus (HERV) activation patterns in immune cells can differentiate between Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS), Fibromyalgia (FM), and co-diagnosed conditions. These HERV profiles correlated with specific immune cell changes and patient symptom severity, suggesting HERVs as potential biomarkers. The co-diagnosed group notably presented HERV profiles closer to healthy controls, potentially indicating a new disease entity.

This study aimed to objectify autonomic and neuropathic complaints in long-COVID patients using electrophysiological tests, but only 14 patients underwent the full detailed evaluation. They found that over a third of long-COVID patients experienced these symptoms, often associated with fatigue and headache, and identified abnormal cutaneous silent period (CSP) suppression but normal sympathetic skin responses (SSR) in the small group studied. Neuropathic complaints were commonly patchy and proximal.

This paper presents MotionTrans, a framework allowing robots to learn complex manipulation tasks by observing human demonstrations in virtual reality (VR) and co-training with robot-collected data. The system enables "zero-shot" task completion on real robots for 9 out of 13 human tasks and significantly boosts performance in few-shot fine-tuning scenarios, bridging the human-robot embodiment gap through data transformation and a weighted co-training strategy.

This study, primarily using a mouse knockout model and human heart cell lines, found that suppressing the GADD45A gene leads to severe cardiac problems, including inflammation, fibrosis, and an enlarged heart. Overexpressing GADD45A in human heart cells, conversely, helped prevent these adverse responses, while human patient data showed a correlation between lower GADD45A levels and increased left ventricular mass and fibrosis. The authors suggest that boosting GADD45A activity could be a therapeutic strategy to slow the progression of heart disease.

This study, conducted on isolated rat heart cells and mouse heart tissue, found that mechanical stretch increases the activity of a specific signaling pathway (β2-Adrenoceptor) within heart cells, which in turn boosts their contractility. This stretch-induced boost relies on tiny cell structures called caveolae, as disrupting them prevented the effect. However, these findings are based on non-human, isolated models and their direct applicability to complex human heart conditions remains to be fully explored.

This clinical report from the American Academy of Pediatrics synthesizes current research, concluding that resistance training is a safe and beneficial component of physical activity for children and adolescents when properly supervised and executed. It highlights the positive outcomes on muscular fitness, athletic performance, and injury reduction, while dispelling common misconceptions about its risks. The report provides guidelines for implementing developmentally appropriate resistance training programs.

This paper speculatively correlates increased breast cancer and melanoma incidence with prolonged sleep on metal spring mattresses, attributing it to the mattresses reflecting "body-resonant radiation" from FM radio transmitters. It suggests that countries with higher spring mattress usage show higher cancer rates, but these are observational correlations without direct causal evidence.

This review summarizes current guidelines and emerging treatments for resistant hypertension in chronic kidney disease (CKD), advocating for individualized, evidence-based management. While novel agents and procedures show promise in blood pressure control and renal benefits, the paper highlights significant limitations, including adverse effects, poor long-term adherence, and high costs that may hinder their widespread use.

This study found significant differences in gut bacteria between 25 chronic fatigue syndrome (CFS) patients and 16 healthy controls, including unique bacterial genera in CFS and altered diversity, but critically failed to collect data on prior antibiotic or probiotic use, a major confounder. Certain gut bacteria, like Oscillibacter and Roseburia, were less abundant in CFS patients and correlated with cognitive function, suggesting a potential gut-brain axis link.

This paper reveals that current climate models overestimate Greenland Ice Sheet meltwater runoff because they don't account for significant refreezing within bare glacier ice. Field measurements and numerical modeling show that this refreezing reduces runoff by an estimated 11-17 Gt annually in southwest Greenland, suggesting an overlooked buffer against sea-level rise from ice melt.

This paper introduces FROSTI, a prototype wavefront actuator that uses thermal radiation to precisely correct mirror distortions in next-generation gravitational-wave detectors. It successfully demonstrates the device's ability to provide higher-order wavefront corrections, maintain high-intensity stability, and meet ultra-high vacuum standards, paving the way for significantly improved detector sensitivity. While promising, the system is currently a prototype optimized for a specific mirror type, and higher-power noise testing is still projected.

This correlational study with 199 human participants found a significant link between how the stomach and brain communicate (stomach-brain coupling) and various aspects of mental health. Higher stomach-brain coupling was associated with worse mental health symptoms like anxiety, depression, and stress, and lower well-being and quality of life. The findings suggest a dimensional signature of mental health related to the gut-brain axis.

This randomized crossover trial investigated the acute effects of sucralose, sucrose, and water on brain appetite regulation, peripheral hormones, and hunger in adults across different body weights. It found that while sucrose significantly increased peripheral glucose, insulin, and GLP-1, and reduced hunger compared to water, sucralose did not elicit similar effects. Moreover, sucralose, when compared to sucrose, resulted in higher hunger ratings and distinct hypothalamic brain responses, with effects varying significantly based on an individual's weight status.

This observational study found a correlation between geomagnetic disturbances and an increased incidence of myocardial infarctions, especially in women, in a single Brazilian city. The authors acknowledge that these findings are not conclusive and emphasize that the study highlights a hypothetical significance rather than a definitive cause-effect relationship.

This mouse and cell line study reveals that an imbalance in building blocks for DNA leads to "mistakes" where RNA components are incorporated into mitochondrial DNA (mtDNA). These faulty mtDNAs then leak out of mitochondria, triggering the cGAS-STING pathway and causing inflammation, particularly in aging kidneys and senescent cells, a process that can be reversed by adding back the correct building blocks.

This astrochemistry paper addresses the challenge of poorly constrained molecular binding energies on interstellar dust grains, crucial for understanding complex organic molecule formation. Using the MOPED algorithm, the authors identify specific ice species whose observations, particularly with improved precision, would best constrain these binding energies. The study recommends prioritizing more precise measurements of already detected species like H2O, CO2, and CH3OH, alongside targeting species such as HCN and H2S.

This paper presents a novel unified policy for legged robots that enables joint force and position control without relying on physical force sensors, instead estimating forces from historical robot states. Through reinforcement learning and experiments on quadrupedal and humanoid robots, the policy demonstrates enhanced success rates (approx. 39.5% higher) in contact-rich tasks like wiping and opening cabinets compared to position-only methods. However, the force estimation accuracy can degrade in high-frequency interactions and at the edges of the robot's workspace, and sim-to-real gaps exist.

This prospective longitudinal cohort study on largely non-hospitalized individuals found that multi-organ impairment persisted in 59% of patients with long COVID one year after initial symptoms, impacting their quality of life and ability to work. However, the study's reliance on self-referred participants and lack of pre-COVID baseline data means results may not be fully generalizable and causal links are harder to establish.

This study introduces an innovative non-invasive 7T MRI method to measure microvascular volumetric pulsatility (mvPI) across different brain layers in humans. Researchers found that mvPI is highest at the brain's surface (pial mater) and significantly higher in the deep white matter of older participants compared to young ones, with even higher mvPI in older individuals with hypertension. However, the study involved a relatively small sample size (11 young, 12 older participants), which limits the generalizability of these age-related findings, especially for subgroup analyses.

This large South Korean study suggests an epidemiological association between COVID-19 vaccination and an increased risk of six specific cancer types (thyroid, gastric, colorectal, lung, breast, and prostate) within one year. The observed associations varied by vaccine type, sex, and age, with booster doses also linked to changes in gastric and pancreatic cancer risks, but the study notes these are associations and not definitive causal links.

This survey experiment on American voters found that while sexual assault/harassment allegations generally decrease a candidate's electability, partisanship significantly mitigates this effect, particularly for Republican voters who are less likely to penalize their own party's candidates. The study used a hypothetical candidate and relies on self-reported voting intentions, and its MTurk sample may not perfectly represent the entire American electorate.

This 17-year observational study found that wild chimpanzees frequently consume fermented palm sap, provided by human tapping, that contains up to 6.9% alcohol using leaf-sponges. The findings suggest that ethanol doesn't deter these apes and contribute to understanding the evolutionary history of ethanol consumption in primates, including humans.