Explore the vast range of titles available.

Despite the remarkable performance of generative large language models (LLMs) on abstractive summarization, they face two significant challenges: their considerable size and tendency to hallucinate. Hallucinations are concerning because they erode reliability and raise safety issues. Pruning is a technique that reduces model size by removing redundant weights, enabling more efficient sparse inference. Pruned models yield downstream task performance comparable to the original, making them ideal alternatives when operating on a limited budget. However, the effect that pruning has upon hallucinations in abstractive summarization with LLMs has yet to be explored. In this paper, we provide an extensive empirical study across five summarization datasets, two state-of-the-art pruning methods, and five instruction-tuned LLMs. Surprisingly, we find that hallucinations are less prevalent from pruned LLMs than the original models. Our analysis suggests that pruned models tend to depend more on the source document for summary generation. This leads to a higher lexical overlap between the generated summary and the source document, which could be a reason for the reduction in hallucination risk.

[...]Members of Parliament voted that the bill become an Act. [...]it was an Act with a difference; it could only come into force if the majority of New Zealanders voted for it in a binding referendum, an additional step stipulated by one of the government coalition members, a party incidentally that has now gone west. Nonetheless despite a well-funded anti-euthanasia campaign that included the publication of misleading and distorted versions of the EOLC Act, and statements that were downright false, the referendum passed overwhelmingly, and work can now focus on establishing how the benefits of this legislation can be made available to all New Zealanders: both to the small number who will actually require assisted dying, and to all other New Zealanders who will benefit from the type of engagement that can occur under this legislation.

Over recent decades, developing countries have witnessed a rapid surge in internet connectivity, driven by an upswing in aid projects dedicated to information and communication technology (ICT) development. While many expect the increased internet connectivity to contribute to the freedom of expression and freer flow of information, others worry that it would rather enable recipient governments to expand their information control and digital surveillance. How do communication aid projects affect internet freedom in recipient countries? Through a cross-national analysis of aid and internet freedom in developing countries, our research reveals that the effect of communication aid on internet freedom varies by the sources of aid projects. Specifically, Chinese aid geared toward communication development has a negative relationship with internet freedom in recipient countries. Conversely, communication aid projects funded by the World Bank and the United States do not exhibit a similarly negative correlation. This paper also finds that the negative association between Chinese communication aid projects and internet freedom is more pronounced in autocratic recipient countries.

The aim of the study is to determine whether the selective aldosterone-receptor antagonist eplerenone delays onset of left ventricular (LV) systolic dysfunction or reduces LV hypertrophy in asymptomatic patients with moderate to severe aortic stenosis. Effects of eplerenone on LV diastolic function and progression of valve stenosis were also evaluated. Sixty-five asymptomatic patients with a peak aortic valve velocity >3.0 m/s and normal LV function were randomized double blind to eplerenone, 100 mg daily (n = 33), or placebo (n = 32) for a median of 19 (interquartile range 15 to 25) months. Cardiac magnetic resonance imaging and echocardiography were performed and N-terminal pro-brain natriuretic peptide was measured at baseline and follow-up. Symptomatic deterioration occurred in 13 subjects randomized to eplerenone and 11 to placebo ( P = .34). Change in LV mass index (mean change ± SD −0.3 ± 14.6 vs +5.1 ± 15 g/m 2 per year, P = .3), LV ejection fraction (+0.0% ± 5.7% vs +0.8% ± 5.7% per year, P = .9), and LV end-systolic volume index (−1.2 ± 9 vs +0.04 ± 12 mL/m 2 per year, P = .8) were small and similar for patients randomized to eplerenone and placebo, respectively. Decrease of aortic valve area (−0.11 ± 0.22 vs −0.18 ± 0.24 cm 2/y, P = .2), worsening of LV diastolic dysfunction by echo-Doppler (E/E' +0.49 ± 0.7 vs +1.32 ± 2.0/year, P = .4), increase in the plasma level of N-terminal pro-brain natriuretic peptide (+63% vs +12% per year, P = .1), and decline in physical function score (9 ± 34 vs 12 ± 37/year, P = .7) were similar for subjects randomized to eplerenone and placebo, respectively. In asymptomatic patients with moderate-severe aortic stenosis, eplerenone did not slow onset of LV systolic or diastolic dysfunction, decrease LV mass, or reduce progression of valve stenosis.

Cardiac diverticula represent an extremely rare but serious cause of cardiac morbidity and mortality. They can result to arrhythmia, sudden cardiac death and ventricular dysfunction but may have no pathological implications. Here is a case of a 60-year-old Maori farmer with both rheumatic mitral valve disease and left ventricular (LV) septal diverticulum. The requirement for mitral valve replacement raised the complex decision of whether to undergo concurrent diverticulum repair. The haemodynamic significance of the diverticulum was impossible to ascertain, although we could not in good conscience leave such a large diverticulum with potential to influence further systolic deterioration. Three months after the procedures, the patient developed severe tricuspid regurgitation which is a first reported association postseptal diverticulum repair. The case highlights that careful consideration is required in repairing LV septal diverticula and an emphasis should be placed on complications and the requirement for repeat surgery during the consent process.

In this dissertation, I study an under-appreciated reality of international aid. It has been long recognized that foreign aid is a multipurpose tool of foreign policy that wealthy countries wield to effect their designs in international politics. What is less often considered is that as countries use foreign aid to accomplish various foreign policy objectives, sometimes these goals may conflict with or support the objectives of other donors. Under the rationalist framework often adopted in the study of international aid, this fact implies strategic interdependence in the aid allocation decisions of donor governments. What are the implications of strategic interdependence in foreign aid, and how does it shape the distribution of economic assistance in developing countries? In practice, this reality has a straightforward implication. While many studies examine patterns in international aid to draw inferences about the goals donors value in their foreign policies, how donors actually distribute foreign aid across recipients does not perfectly reflect their values. To the contrary, strategic interdependence creates incentives for donor governments deviate from giving aid purely in accordance with their objectives. Without a grasp for the role that conflicting and common foreign policy goals play in the allocation of international aid, scholars risk mis-identifying relationships between factors thought to determine aid flows and where donor governments offer more or less economic assistance. While previous studies have addressed this issue in some shape or fashion, none adopt a theoretical perspective or an empirical strategy as comprehensive as I do in this dissertation. Taken together, the four chapters that follow make contributions to theory, measurement, and analysis, providing novel results about the implications of strategic interdependence in international aid. In Chapter 1, I develop a mathematical model to study the implications of strategic interdependence in the political economy of aid. Analysis of the model underscores the mechanisms that drive strategic interdependence and reveals why donor governments may under or over commit resources in developing countries in pursuit of their foreign policy goals. It specifically identifies the conditions under which empirical analysis will provide informative estimates of donor responses to the giving of others. It also yields predictions about how the comparative resource endowments of donor governments and the strategic valence of donor goals in developing countries push some donors to the top, and others to the bottom, in committing aid in recipients. The model also offers insight into the welfare implications of strategic interdependence and suggests mechanisms that may account for the repeated failure of donor countries to successfully collaborate in the allocation of aid. Analysis shows that an uncoordinated Nash equilibrium among donor governments can often have an unintiutive location relative to a Pareto improving alternative under collective optimization. Even more, the existence of a Pareto improving alternative is not guaranteed. In many instances, the adoption of a collectively optimal solution may be individually worse for at least one donor government relative to a Nash equilibrium. These results illustrate the kinds of stumbling blocks that may continue to impede donor collaboration and, thus, the kinds of factors that ought to be considered in the design of institutions.In Chapter 2, I develop novel measures that will assist in empirical analysis of strategic interdependence in international aid. Using an approach I call SSC (sum of the squared covariances), I create composite measures of interest-based and needs-based factors that drive donor giving in developing countries. The first, which I call donor interest (DI), captures material, strategic, geographic, and social ties between donor governments and developing countries that make the latter salient targets for foreign aid to the former. The second measure, which I call recipient need (RN), captures developing country characteristics linked to greater need for donor assistance. By creating composite measures, I am able to probe how the interaction of donor-recipient ties and depth of recipient development need interact to shape donor responses to the aid given by others in developing countries with greater parsimony. To validate SSC, I compare SSC-constructed versions of DI and RN to those created by alternative methods of measurement construction and use various modeling techniques to compare their prognostic power. The SSC derived measures consistently outperform those created via other approaches. Further, to assess what these measures can reveal about broad patterns in international aid, I perform pooled analyses and analyses for individual donor governments and years to compare and contrast donor responsiveness to interest-based and needs-based factors. The patterns in some cases align with conventional wisdom about donor motives, but in other cases support different inferences about donor motives than reported in other studies.In Chapter 3, using the DI and RN measures created in the previous chapter, I provide confirmatory evidence that donor governments do in fact make their aid allocation on the basis of the giving of others and that these responses vary systematically with respect to the strength of interest-based and needs-based drivers of giving. Across the majority of the parameter space, donors show deference to others by giving less aid in recipients where others give more. But, there are instances where we observe interesting deviations from this response. On average, a donor gives more aid in recipients where others give more either when needs-based factors are low and interest-based factors for a donor are high, or when needs-based factors are high and interest-based factors are low. Conversely, a donor gives less aid in recipients were others give more either when needs-based factors are high and interest-based factors are high, or when needs-based factors are low and interest-based factors are low. Though many explanations may account for these patterns, they are arguably consistent with two prominent perspectives in the aid literature. One of these is called “targeted development” and the other “aid-for-policy exchange.” The former is consistent with a common interest among donors for promoting development as a way to prevent transnational spillovers of developing country problems from affecting industrialized country populations. The latter is consistent with rival interests over policy concessions from recipients. Targeted development is proposed to dominate at the confluence of high interest and high need, while aid-for-policy exchange is proposed to dominate at low levels of interest and high need or high levels of interest and low need.Beyond the specifics of donor responses, the analysis in Chapter 3 further shows that strategic interdependence does indeed lead to deviations between what donors value and how they actually distribute aid. A comparison of estimates on DI and RN with and without accounting for different responses to peer aid reveals striking differences in how interest-based and need-based factors determine patterns in aid giving. When strategic interdependence is not accounted for, regression estimates on DI and RN are attenuated in value, while second-order interactions between DI and RN are either suppressed or inflated depending on the sample of years under analysis.Finally, in Chapter 4, I test arguments about how strategic interdependence shapes when and where donor governments come to be the dominant source of foreign aid in developing countries. Building on the theoretical analysis in Chapter 1 and insights on how donors respond to other-donor aid in different contexts from Chapter 3, we should expect a donor government to hold lead donor status at the confluence of two sets of conditions. A donor should take the lead, first, in high need recipients where interest-based factors are comparatively stronger for said donor relative to others, and second, when a donor is better resourced than its peers. The first set of conditions implies a development focus for donors (a non-rival goal) while the second confers a material advantage in financing international aid. Building on the dataset used in Chapter 3, I construct a measure of lead donorship and find patterns in donor giving consistent with the above argument. I further take a closer look at variations in lead donorship in Latin America and Southeast Asia where I find some patterns that are consistent with proposed theoretical mechanisms, and others that suggest additional dynamics are at work as donors give foreign aid. Overall, the results in Chapter 4 reflect the first attempt to empirically model the determinants of lead donorship. The patterns I uncover have significance, not only for the support they provide for my theoretical argument, but also for informing research on the consequences of lead donorship for aid effectiveness in developing countries.

Addresses what the authors see as concerning, confusing, conflating, irrelevant and/or inaccurate themes in the Report on Euthanasia for the New Zealand Medical Association by Grant Gillett, especially in relation to voluntary assisted dying (VAD) and hospice/palliative care (H/PC). Source: National Library of New Zealand Te Puna Matauranga o Aotearoa, licensed by the Department of Internal Affairs for re-use under the Creative Commons Attribution 3.0 New Zealand Licence.

Abstract Objective: To determine the proportion of patients presenting with acute myocardial infarction who are eligible for thrombolytic therapy. Design: Cohort follow up study. Setting: The four coronary care units in Auckland, New Zealand. Subjects: All 3014 patients presenting to the units with suspected myocardial infarction in 1993. Main outcome measures: Eligibility for reperfusion with thrombolytic therapy (presentation within 12 hours of the onset of ischaemic chest pain with ST elevation >/=2 mm in leads V1-V3, ST elevation >/=1 mm in any other two contiguous leads, or new left bundle branch block); proportions of (a) patients eligible for reperfusion and (b) patients with contraindications to thrombolysis; death (including causes); definite myocardial infarction. Results: 948 patients had definite myocardial infarction, 124 probable myocardial infarction, and nine ST elevation but no infarction; 1274 patients had unstable angina and 659 chest pain of other causes. Of patients with definite or probable myocardial infarction, 576 (53.3%) were eligible for reperfusion, 39 had definite contraindications to thrombolysis (risk of bleeding). Hence 49.7% of patients (537/1081) were eligible for thrombolysis and 43.5% (470) received this treatment. Hospital mortality among patients eligible for reperfusion was 11.7% (55/470 cases) among those who received thrombolysis and 17.0% (18/106) among those who did not. Conclusions: On current criteria about half of patients admitted to coronary care units with definite or probable myocardial infarction are eligible for thrombolytic therapy. Few eligible patients have definite contraindications to thrombolytic therapy. Mortality for all community admissions for myocardial infarction remains high. Key messages Less than 10% of patients eligible for reperfusion have contraindications to thrombolysis The hospital mortality for all patients with acute myocardial infarction remains high (14%) Better treatments are required to reduce mortality in both reperfusion eligible and reperfusion ineligible patients

Quantization and pruning form the foundation of compression for neural networks, enabling efficient inference for large language models (LLMs). Recently, various quantization and pruning techniques have demonstrated remarkable performance in a post-training setting. They rely upon calibration data, a small set of unlabeled examples that are used to generate layer activations. However, no prior work has systematically investigated how the calibration data impacts the effectiveness of model compression methods. In this paper, we present the first extensive empirical study on the effect of calibration data upon LLM performance. We trial a variety of quantization and pruning methods, datasets, tasks, and models. Surprisingly, we find substantial variations in downstream task performance, contrasting existing work that suggests a greater level of robustness to the calibration data. Finally, we make a series of recommendations for the effective use of calibration data in LLM quantization and pruning.

Summarises and comments on the findings of the ‘End of Life Choice’ survey of general practitioners (GP) by Horizon Research evaluating GPs' support for aspects of the current End of Life Choice Bill (EOLC Bill) being considered by the Justice Select Committee. Refers to a similar 2017 survey on doctors' support of medical aid in dying (MAID). Source: National Library of New Zealand Te Puna Matauranga o Aotearoa, licensed by the Department of Internal Affairs for re-use under the Creative Commons Attribution 3.0 New Zealand Licence.