Publications

Playing video games requires perception, memory, and planning, exactly the faculties modern large language model (LLM) agents are expected to master. We study the major challenges in using popular video games to evaluate modern LLMs and find that directly dropping LLMs into games cannot make an effective evaluation, for three reasons -- brittle vision perception, prompt sensitivity, and potential data contamination. We introduce lmgame-Bench to turn games into reliable evaluations. lmgame-Bench features a suite of platformer, puzzle, and narrative games delivered through a unified Gym-style API and paired with lightweight perception and memory scaffolds, and is designed to stabilize prompt variance and remove contamination. Across 13 leading models, we show lmgame-Bench is challenging while still separating models well. Correlation analysis shows that every game probes a unique blend of capabilities often tested in isolation elsewhere. More interestingly, performing reinforcement learning on a single game from lmgame-Bench transfers both to unseen games and to external planning tasks. Our evaluation code is available at https://github.com/lmgame-org/GamingAgent/lmgame-bench.

An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures.

A new type of metallic electromagnetic structure has been developed that is characterized by having high surface impedance. Although it is made of continuous metal, and conducts dc currents, it does not conduct ac currents within a forbidden frequency band. Unlike normal conductors, this new surface does not support propagating surface waves, and its image currents are not phase reversed. The geometry is analogous to a corrugated metal surface in which the corrugations have been folded up into lumped-circuit elements, and distributed in a two-dimensional lattice. The surface can be described using solid-state band theory concepts, even though the periodicity is much less than the free-space wavelength. This unique material is applicable to a variety of electromagnetic problems, including new kinds of low-profile antennas.

Abstract The selective functionalization of one C−H bond over others in nearly identical steric and electronic environments can facilitate the construction of complex molecules. We report site‐selective functionalizations of C−H bonds, differentiated solely by remote substituents, catalyzed by artificial metalloenzymes (ArMs) that are generated from the combination of an evolvable P450 scaffold and an iridium‐porphyrin cofactor. The generated systems catalyze the insertion of carbenes into the C−H bonds of a range of phthalan derivatives containing substituents that render the two methylene positions in each phthalan inequivalent. These reactions occur with site‐selectivity ratios of up to 17.8:1 and, in most cases, with pairs of enzyme mutants that preferentially form each of the two constitutional isomers. This study demonstrates the potential of abiotic reactions catalyzed by metalloenzymes to functionalize C−H bonds with site selectivity that is difficult to achieve with small‐molecule catalysts.

Recyclable catalysts, especially those that display selective reactivity, are vital for the development of sustainable chemical processes. Among available catalyst platforms, heterogeneous catalysts are particularly well-disposed toward separation from the reaction mixture via filtration methods, which renders them readily recyclable. Furthermore, heterogeneous catalysts offer numerous handles-some without homogeneous analogues-for performance and selectivity optimization. These handles include nanoparticle size, pore profile of porous supports, surface ligands and interface with oxide supports, and flow rate through a solid catalyst bed. Despite these available handles, however, conventional heterogeneous catalysts are themselves often structurally heterogeneous compared to homogeneous catalysts, which complicates efforts to optimize and expand the scope of their reactivity and selectivity. Ongoing efforts in our laboratories are aimed to address the above challenge by heterogenizing homogeneous catalysts, which can be defined as the modification of homogeneous catalysts to render them in a separable (solid) phase from the starting materials and products. Specifically, we grow the small nanoclusters in dendrimers, a class of uniform polymers with the connectivity of fractal trees and generally radial symmetry. Thanks to their dense multivalency, shape persistence, and structural uniformity, dendrimers have proven to be versatile scaffolds for the synthesis and stabilization of small nanoclusters. Then these dendrimer-encapsulated metal clusters (DEMCs) are adsorbed onto mesoporous silica. Through this method, we have achieved selective transformations that had been challenging to accomplish in a heterogeneous setting, e.g., π-bond activation and aldol reactions. Extensive investigation into the catalytic systems under reaction conditions allowed us to correlate the structural features (e.g., oxidation states) of the catalysts and their activity. Moreover, we have demonstrated that supported DEMCs are also excellent catalysts for typical heterogeneous reactions, including hydrogenation and alkane isomerization. Critically, these investigations also confirmed that the supported DEMCs are heterogeneous and stable against leaching. Catalysts optimization is achieved through the modulation of various parameters. The clusters are oxidized (e.g., with PhICl₂) or reduced (e.g., with H₂) in situ. Changing the dendrimer properties (e.g., generation, terminal functional groups) is analogous to ligand modification in homogeneous catalysts, which affect both catalytic activity and selectivity. Similarly, pore size of the support is another factor in determining product distribution. In a flow reactor, the flow rate is adjusted to control the residence time of the starting material and intermediates, and thus the final product selectivity. Our approach to heterogeneous catalysis affords various advantages: (1) the catalyst system can tap into the reactivity typical to homogeneous catalysts, which conventional heterogeneous catalysts could not achieve; (2) unlike most homogeneous catalysts with comparable performance, the heterogenized homogeneous catalysts can be recycled; (3) improved activity or selectivity compared to conventional homogeneous catalysts is possible because of uniquely heterogeneous parameters for optimization. In this Account, we will briefly introduce metal clusters and describe the synthesis and characterizations of supported DEMCs. We will present the catalysis studies of supported DEMCs in both the batch and flow modes. Lastly, we will summarize the current state of heterogenizing homogeneous catalysis and provide future directions for this area of research.

This study aimed to examine changes in depression and anxiety symptoms from before to during the first 6 months of the COVID-19 pandemic in a sample of 1,339 adolescents (9-18 years old, 59% female) from three countries. We also examined if age, race/ethnicity, disease burden, or strictness of government restrictions moderated change in symptoms. Data from 12 longitudinal studies (10 U.S., 1 Netherlands, 1 Peru) were combined. Linear mixed effect models showed that depression, but not anxiety, symptoms increased significantly (median increase = 28%). The most negative mental health impacts were reported by multiracial adolescents and those under 'lockdown' restrictions. Policy makers need to consider these impacts by investing in ways to support adolescents' mental health during the pandemic.

Abstract Purpose This study sought to identify the organizational factors associated with team and network effectiveness of the Athena Breast Health Network, a multi-site collaboration between five University of California health systems. Design/methodology/approach Providers, managers, and support staff completed self-administered surveys over three years. Statistical analyses at the network and medical center levels tested hypotheses regarding the correlates of effective teams and perceived network effectiveness over time. Findings Perceived team effectiveness was positively correlated with group culture and environments which support collaboration, negatively correlated with hierarchical culture, and negatively associated with professional tenure at year two. As measured by increasing team effectiveness scores over time and Athena’s potential impact on patient care, perceived network effectiveness was positively associated with team effectiveness. Research limitations/implications Results do not allow us to conclude that a certain type of culture “causes” team effectiveness or that team effectiveness “causes” greater perceptions of progress over time. Subsequent studies should examine these variables simultaneously. Further research is needed to examine the role of payment incentives, internal reward systems, the use of electronic health records, public disclosure of performance data, and depth of leadership within each organization and within the network overall. Practical implications Focusing on group affiliation and participation may improve team member perceptions regarding effectiveness and impact on patient care. Originality/value Relatively little is known about the adaptive processes that occur within inter-organizational networks to achieve desired goals, and particularly the roles played by multi-disciplinary inter-professional teams. We studied a network comprising multiple campuses actively involved in better understanding, preventing, and treating a complex disease.

An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures.

Abstract Both innate and adaptive immune systems contribute to host defense against infection with Mycobacterium tuberculosis. NK cells have been associated with early resistance against intracellular pathogens and are known to be potent producers of the cytokine IFN-γ. In C57BL/6 mice infected by aerosol exposure with M. tuberculosis, NK cells increased in the lungs over the first 21 days of infection. Expansion of the NK cell subset was associated with increased expression of activation and maturation markers. In addition, NK cells isolated from the infected lungs were capable of producing IFN-γ and became positive for perforin. In vivo depletion of NK cells using a lytic Ab had no influence on bacterial load within the lungs. These findings indicate that NK cells can become activated during the early response to pulmonary tuberculosis in the mouse model and are a source of IFN-γ, but their removal does not substantially alter the expression of host resistance.