Search for: All records

Navigating dilemmas involving conflicting values is challenging even for humans in high-stakes domains, let alone for AI, yet prior work has been limited to everyday scenarios. To close this gap, we introduce CLASH (Character perspective-based LLM Assessments in Situations with High-stakes), a meticulously curated dataset consisting of 345 high-impact dilemmas along with 3,795 individual perspectives of diverse values. CLASH enables the study of critical yet underexplored aspects of value-based decision-making processes, including understanding of decision ambivalence and psychological discomfort as well as capturing the temporal shifts of values in the perspectives of characters. By benchmarking 14 non-thinking and thinking models, we uncover several key findings. (1) Even strong proprietary models, such as GPT-5 and Claude-4-Sonnet, struggle with ambivalent decisions, achieving only 24.06 and 51.01 accuracy. (2) Although LLMs reasonably predict psychological discomfort, they do not adequately comprehend perspectives involving value shifts. (3) Cognitive behaviors that are effective in the math-solving and game strategy domains do not transfer to value reasoning. Instead, new failure patterns emerge, including early commitment and overcommitment. (4) The steerability of LLMs towards a given value is significantly correlated with their value preferences. (5) Finally, LLMs exhibit greater steerability when reasoning from a third-party perspective, although certain values (e.g., safety) benefit uniquely from first-person framing. 

The rapid development of adeno-associated viral vectors (AAV) to treat genetic disease has placed increased emphasis on the design of efficient downstream manufacturing processes. This study investigated the potential of using single pass tangential flow filtration (SPTFF) as a novel means of concentrating and purifying AAV clarified cell lysate (CCL). AAV stability studies revealed the shear-sensitive nature of the AAV capsids, with evidence of aggregation and fragmentation following repeated passages through a peristaltic pump (as would occur during batch ultrafiltration). SPTFF experiments focused on first identifying the membrane(s) that permitted high yield of AAV (negligible sieving into the permeate) along with substantial host cell protein (HCP) removal. Experiments were then performed at various permeate fluxes, which revealed that stable SPTFF processes can be achieved by operating below a critical flux for fouling (Jfoul). 300 kDa regenerated cellulose (RC) membranes were identified as optimal for this application, given their ability to provide complete AAV retention with high removal of HCP (>90%) when operated below Jfoul. The critical flux during SPTFF was increased by preconditioning the CCL through a positively-charged adsorptive filter, which reduced the concentration of foulants prior to SPTFF. These studies provide the first demonstration of SPTFF for the concentration and purification of AAV clarified cell lysate while minimizing shear exposure. 

Abstract Accurate in-hospital length of stay prediction is a vital quality metric for hospital leaders and health policy decision-makers. It assists with decision-making and informs hospital operations involving factors such as patient flow, elective cases, and human resources allocation, while also informing quality of care and risk considerations. The aim of the research reported in this paper is to use survival analysis to model General Internal Medicine (GIM) length of stay, and to use Shapley value to support interpretation of the resulting model. Survival analysis aims to predict the time until a specific event occurs. In our study, we predict the duration from patient admission to discharge to home, i.e., in-hospital length of stay. In addition to discussing the modeling results, we also talk about how survival analysis of hospital length of stay can be used to guide improvements in the efficiency of hospital operations and support the development of quality metrics. 

Abstract We prove lower bounds on the density of regular minimal cones of dimension less than seven provided the complements of the cones are topologically nontrivial. 

Abstract Most seasonal and pandemic influenza vaccines are derived from inactivated or attenuated virus propagated in chicken eggs, while more advanced delivery technologies, such as the use of recombinant proteins and adjuvants, are under‐utilized. In this study, the E2 protein nanoparticle (NP) platform is engineered to synthesize vaccines that simultaneously co‐deliver influenza hemagglutinin (H5) antigen, TLR5 agonist flagellin (FliCc), and TLR9 agonist CpG 1826 (CpG) all on one particle (termed H5‐FliCc‐CpG‐E2), with uniform molecular orientation significant for immunomodulation. Antigen‐bound NP formulations elicit higher IgG antibody responses and broader homosubtypic cross‐reactivity against different H5 variants than unconjugated antigen alone. IgG1/IgG2c skewing is modulated by adjuvant type and NP attachment. Conjugation of flagellin to the NP causes significant IgG1 (Th2) skewing while attachment of CpG yields significant IgG2c (Th1) skewing, and simultaneous conjugation of both flagellin and CpG results in a balanced IgG1/IgG2c (Th2/Th1) response. Animals immunized with E2‐based NP vaccines and subsequently challenged with H5N1 influenza show 100% survival, and only animals that receive adjuvanted NP formulations are also protected against morbidity. This investigation highlights that NP‐based delivery of antigen and multiple adjuvants can be designed to effectively modulate the strength, breadth toward variants, and bias of an immune response against influenza viruses.