Search

This study examines the capabilities of advanced Large Language Models (LLMs), particularly the o1 model, in the context of literary analysis. The outputs of these models are compared directly to those produced by graduate-level human participants. By focusing on two Nobel Prize-winning short stories, 'Nine Chapters' by Han Kang, the 2024 laureate, and 'Friendship' by Jon Fosse, the 2023 laureate, the research explores the extent to which AI can engage with complex literary elements such as thematic analysis, intertextuality, cultural and historical contexts, linguistic and structural innovations, and character development. Given the Nobel Prize's prestige and its emphasis on cultural, historical, and linguistic richness, applying LLMs to these works provides a deeper understanding of both human and AI approaches to interpretation. The study uses qualitative and quantitative evaluations of coherence, creativity, and fidelity to the text, revealing the strengths and limitations of AI in tasks typically reserved for human expertise. While LLMs demonstrate strong analytical capabilities, particularly in structured tasks, they often fall short in emotional nuance and coherence, areas where human interpretation excels. This research underscores the potential for human-AI collaboration in the humanities, opening new opportunities in literary studies and beyond.

In this paper, we shall prove that, on a non-flat Riemannian vector bundle over a compact Riemannian manifold, the smooth solution of the Yang-Mills flow will blow up in finite time if the energy of the initial connection is small enough. We also consider the finite time blow up for the Yang-Mills flow with the initial curvature near the harmonic form. Furthermore, when $E$ is a holomorphic vector bundle over a compact K\"ahler manifold, then $E$ will admit a projective flat structure if the trace free part of Chern curvature is small enough.

We add to recent evidence on deindustrialization and document a new pattern: increasing industry polarization over time. We assess whether these new features of structural change can be explained by a dynamic open economy model with two primary driving forces, sector-biased productivity growth and sectoral trade integration. We calibrate the model to the same countries used to document our patterns. We find that sector-biased productivity growth is important for deindustrialization by reducing the relative price of manufacturing to services, and sectoral trade integration is important for industry polarization through increased specialization. The interaction of these two driving forces is also essential as increased trade openness transmits global technological change to each country's relative prices, sectoral specialization, and sectoral trade imbalances. [ABSTRACT FROM AUTHOR]

The active power filter (APF) is attracting more and more attention for its outstanding performance in current and voltage ripple compensation. As modern high-energy accelerators are demanding much more stringent current ripple guideline, the APF is introduced to the magnet power supply (MPS) in accelerator system. However, the conventional APF has a lot of shortages and drawbacks due to its traditional topology, such as complex structure, nonadjustable working voltage, requirement of power supply, and so on. This paper proposes a new topology of APF, which is working as two types of chopper circuits. This APF need not extra electricity, but to use the power of the MPS current ripple to realize ripple depressing. At the end of this paper, the experiment result proves its feasibility and effect., Comment: 9 pages, 11 figures

We prove the existence and uniqueness of solution of the obstacle problem for quasilinear Stochastic PDEs with non-homogeneous second order operator. Our method is based on analytical technics coming from the parabolic potential theory. The solution is expressed as a pair $(u,\nu)$ where $u$ is a predictable continuous process which takes values in a proper Sobolev space and $\nu$ is a random regular measure satisfying minimal Skohorod condition. Moreover, we establish a maximum principle for local solutions of such class of stochastic PDEs. The proofs are based on a version of It\^o's formula and estimates for the positive part of a local solution which is non-positive on the lateral boundary., Comment: 19 pages. arXiv admin note: substantial text overlap with arXiv:1202.3296, arXiv:1210.3445, arXiv:1201.1092

We prove a maximum principle for local solutions of quasilinear stochastic PDEs with obstacle (in short OSPDE). The proofs are based on a version of It\^o's formula and estimates for the positive part of a local solution which is non-positive on the lateral boundary.

Background: Counts are familiar outcomes in education research settings, including those involving tests of interventions. Clustered data commonly occur in education research studies, given that data are often collected from students within classrooms or schools. There is a wide array of distributions and models that can be used for clustered count data and that adequately capture the shape and nature of diverse count distributions (Cameron & Trivedi, 2013; Grimm & Stegmann, 2019; Hilbe, 2011). In this paper we focus on two specific models for multilevel analyses of counts that are forms of generalized linear mixed models (GLMM) for discrete data: multilevel Poisson (P) and multilevel negative binomial (NB). We cover theoretical background of both approaches, and present a series of guiding questions to assist researchers in determining which may be the most appropriate model for their particular analyses and research questions. Purpose/objective: Our paper focuses on two comparative applications (P and NB) of multilevel models for counts. Our objectives are to (1) provide theoretical background; (2) clarify assessment of degree of clustering for each model, as well as parameter interpretation; (3) review guiding questions for researchers to assist in model comparison and choice; (4) highlight software issues likely encountered when analyzing clustered count data; and (5) briefly discuss possible extensions or other analysis options. Application: Our motivating example is based on the ECLS-K (https://nces.ed.gov/ecls/kindergarten.asp) and was simulated/modified for demonstration purposes. These data correspond to a study in which proficiency counts were collected for a sample of n = 12,938 kindergarten children sampled from within J = 720 early-grade schools. Proficiency (profcount) was scored based on the number of correct responses to a set of twelve early reading and early math skills items, with the total possible count ranging from 0 to 12 (M = 2.05 items, S = 1.85, S[superscript 2] = 3.44). Variables used to predict proficiency at the child level included male (dichotomous, 50% male), highest level of mother's education ("c_momed," centered and continuous, Md = 5 (at least some college)), and food insecurity ("foodinsec," dichotomous, 8% experiencing food insecurity). Predictors of pro ciency considered at the school level included "public" (78% public schools) and count of crime/conflict issues in the surrounding community ranging from 0 to 21 ("nbhoodprobs," M = 6.29 items, S = 3.39, S[superscript 2] = 11.46). Methods: The Poisson distribution makes an assumption of equidispersion which implies that the mean of the outcome (conditional on covariates in the model) is equal to its variance: E(y[subscript 1]) = [subscript lambda i] and V(y[subscript 1]) = [subscript lambda i]. Equidispersion is violated when the variance exceeds the mean, commonly known as overdispersion. The NB distribution we discuss here -- also referred to as NB2 (Cameron & Trivedi, 2013; Hilbe, 2011) relaxes this assumption through the use of a Poisson-Gamma mixture which captures overdispersion in the counts. As a Poisson-Gamma mixture, the NB model has the same mean structure as the Poisson, with conditional variance, [lambda subscript i] + [lambda subscript i superscript 2]/[theta]. The extra dispersion term tends to 0 as [theta] tends to infinity, representing an indirect relationship between [theta] and dispersion (Friendly & Myer, 2016; Hilbe, 2011). To capture a direct association between the parameter contributing to the increased dispersion and the variance, the negative binomial model can be parameterized so that [alpha] = 1/[theta] represents the factor by which the second term is increased; thus, this parameterization for the Poisson-Gamma mixture has variance represented by [lambda] + [alpha][lambda superscript 2] . As [alpha] increases so does the additional variance, and as [alpha] tends towards 0, the Gamma distribution contributes no additional dispersion and the variance of the Poisson-Gamma mixture converges to that of the Poisson. Interest focuses on estimating [alpha], which is called the heterogeneity or NB dispersion parameter under this parameterization (Fox, 2017; Hilbe, 2011), and assessing whether its inclusion provides better fit for the model to the data over that of the Poisson. A multilevel approach accounts for variability that arises from clustered or correlated count data. Degree of clustering is an important issue for multilevel models in general, and we present recent advances in variance partitioning for multilevel count models including estimation of the intraclass correlation coefficient (ICC) following Leckie et al. (2020). We propose four guiding questions for comparison of models and decisions on multilevel P versus NB analyses: (1) To what degree is clustering present in the data, and does accounting for clustering improve model fit? (2) After adjusting for clustering, is the assumption of equidispersion reasonable, or should this assumption be relaxed? (3) Does inclusion of covariates improve model fit, and should any level-one predictors be treated as random or fixed? and (4) How are parameter estimates interpreted for the final model (multilevel P or NB)? Results: We focus on specific aspects to the guiding questions here, and full description is included in the paper and our presentation. Selected model results for our demonstration data are presented in Table 1 for two-level random-intercept P and NB analyses. Variance partitioning based on Leckie, et al. (2020) yields ICC(P) = 0.37, and ICC(NB) = 0.28. Once clustering is accommodated, itself a source of overdispersion, the NB model including extra dispersion as a random effect at level-1 indicates somewhat lower estimated correlation between students from the same school. The extra dispersion parameter for the NB model is statistically different from 0, (Wald's Z = 0.15/0.01 = 15.0) and thus the equidispersion assumption is not upheld, even after all predictors are included in the model (0.12/0.01 = 12.0). Comparing the full NB model to the corresponding nested P model via likelihood ratio test, X[superscript 2 subscript 1] = 211.61, p < 0.0001, the NB model is preferred; this model also has the smallest AIC of the models presented. Summary/Discussion: Multilevel NB models are becoming more common in education research and in the literature. The four questions posed here can help guide education researchers in their choice of multilevel count models particularly between P and NB analyses. Our paper and presentation will include further details on the guiding questions and model parameters including interpretation of incidence rate ratios (exp(gamma)).

Motivated by increasing trade and fragmentation of production across countries since World War II, we build a dynamic two-country model featuring sequential, multistage production and capital accumulation. As trade costs decline over time, global-value-chain (GVC) trade expands across countries, particularly more in the faster growing country, consistent with the empirical pattern. The presence of GVC trade boosts capital accumulation and economic growth and magnifies dynamic gains from trade. At the same time, endogenous capital accumulation shapes comparative advantage across countries, impacting the dynamics of GVC trade: a country becoming more capital abundant concentrates more on the capital-intensive stage of the production. [ABSTRACT FROM AUTHOR]

Services, which are less traded than goods, rose from 50 percent of world expenditure in 1970 to 80 percent in 2015. Such structural change restrained "openness"-the ratio of world trade to world GDP-over this period. We quantify this with a general equilibrium trade model featuring non-homothetic preferences and input-output linkages. Openness would have been 70 percent in 2015, 23 percentage points higher than the data, if expenditure patterns were unchanged from 1970. Structural change is critical for estimating the dynamics of trade barriers and welfare gains from trade. Ongoing structural change implies declining openness, even absent rising protectionism. [ABSTRACT FROM AUTHOR]

As part of a study for a proposed hydropower facility, the authors conducted extensive field observations in the Upper Susitna basin, a 13,289 km2 (5,130 mi2) glacierized catchment in central Alaska, in 2012-2014. This comprehensive data set includes meteorological, glacier mass balance, snow cover, and soil measurements. We also include digitized snow depth data from a set of similar observations collected in the 1980s. The data will be useful for hydrological and glaciological studies, including modeling efforts. [ABSTRACT FROM AUTHOR]