Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fine-tuning datasets of thousands or tens of thousands of examples. By contrast, humans can generally perform a new language task from only a few examples or from simple instructions -something which current NLP systems still largely struggle to do. Here we show that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even reaching competitiveness with prior state-of-the-art finetuning approaches. Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks, as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetic. At the same time, we also identify some datasets where GPT-3's few-shot learning still struggles, as well as some datasets where GPT-3 faces methodological issues related to training on large web corpora. Finally, we find that GPT-3 can generate samples of news articles which human evaluators have difficulty distinguishing from articles written by humans. We discuss broader societal impacts of this finding and of GPT-3 in general.

The VA outpatients have substantially worse health status than non-VA populations. Mental health differences between the young and old veterans who use the VA health care system are sharply contrasting; the young veterans are sicker, suggesting substantially higher resource needs. Mental health differences may explain much of the worse health-related quality of life in young veterans. As health care systems continue to undergo a radical transformation, the Department of Veterans Affairs should focus on the provision of mental health services for its younger veteran.

Recently, the Veterans Administration (VA) Under Secretary for Health has designated functional status as one of the domains of value for the system, given its increasing importance for clinical care. The Veterans Health Study (VHS) was designed to assist the VA in monitoring outcomes and measuring the case mix of patients who use the VA. The Veterans SF-36 (short form functional status assessment for veterans) was administered to 2425 veterans receiving ambulatory care. Measures of the Veterans SF-36 were strongly correlated with sociodemographics and morbidities of the veterans. Young veterans had poorer mental health status than older veterans. Veterans who used ambulatory care in the VHS reported lower levels of health status, reflecting more disease than a non-VA civilian population. These measures of health are important indicators of the disease burden or case mix of the patients and are pertinent to health systems such as the VA for resource allocation decisions and as outcomes of care.

Regret was substantial and associated with treatment choice and quality of life. It may derive from underlying psychosocial distress and problematic communication with physicians when decisions are being reached and over subsequent years.

Role functioning and its limitations due to one's health is an important aspect of health-related quality of life (HRQoL). The Medical Outcomes Study (MOS) SF-36 includes 2 role functioning scales: role limitations due to physical health problems (RP) or emotional problems (RE). Although they capture important concepts of HRQoL, these 2 scales have some limitations in their measurement properties. Using dichotomized sets of response choices, the scales are limited in their distributional properties (eg, higher standard deviation than other SF-36 scales) and ability to discriminate between clinically relevant groups. In this study, we ascertain the improvements to these 2 scales using 5-point ordinal response choices for each of the scale items. Two thousand one hundred sixty-two patients from the Veterans Health Study (VHS), an observational study of health outcomes in patients receiving ambulatory care, completed a health status questionnaire and a medical history. The health questionnaire included (1) the MOS SF-36, in which the RP and RE items used dichotomized yes/no responses; and (2) a set of modified RP and RE items that used 5-response choices for each of the items, ranging from "no, none of the time" to "yes, all of the time." We compared the original and modified RP and RE scales using internal consistency reliability and factor analysis. We tested item convergent and discriminant validity using multitrait scaling, and scale discriminant validity using ordinary least squares regression. Results indicate that the modifications to the original RP and RE scales accomplish important gains in the distributional properties of the scales. The floor and ceiling effects of the 2 scales have been reduced and the reliability of the RP scale has increased (0.87-0.95). Factor analysis and multitrait scaling tests indicate that the modified items have the same interpretation as the original items. Tests of discriminant validity indicate that the modified RP and RE scales have greater explanatory power for measures of disease burden, depression, and disease severity. The modified SF-36 role scales are clearly superior to the original versions. The modifications have increased the explained variability, suggesting greater explanatory power and more information obtained by the role functioning measures. The modified RP and RE are capturing a wider spectrum of disease severity, in part due to the lowering of the floor and raising of the ceiling of the scales. Additional work needs to test these improvements in other populations and to expand the analysis to track the responsiveness of the modified scales to clinically and socially important changes over time.

Pretreatment function and the primary treatment modality for early stage PC strongly predict the affected organ systems and time course of dysfunction. With this information, patients and their physicians may refine their choice of treatment and better anticipate its consequences.

WHAT'S KNOWN ON THIS SUBJECT: Rates of human papillomavirus (HPV) vaccination lag behind other adolescent vaccines. Research indicates that provider recommendation is the key to improving HPV vaccination rates and that most adolescents who are unvaccinated received other vaccines, indicating missed opportunities for HPV vaccination. WHAT THIS STUDY ADDS:This study explores in-depth the content of provider-patient conversations that either create or prevent opportunities for HPV vaccination. Effective and ineffective conversations are presented with the goal of providing practical tools to improve communication regarding HPV vaccines.abstract OBJECTIVE: The goal of this study was to identify the rationale by parents/guardians and providers for delaying or administering human papillomavirus (HPV) vaccination to girls. METHODS:Qualitative interviews were conducted with parents/ guardians accompanying their vaccine-eligible 11-to 17-year-old daughters to medical visits. Interviews were conducted in 1 public clinic and 3 private practice settings to ascertain why girls did or did not receive HPV vaccination. Questions probed vaccine decisionmaking from the point of view of parents/guardians and providers.RESULTS: A total of 124 parents/guardians and 37 providers participated. The most common reasons parents reported for not vaccinating their daughters was the lack of a physician recommendation (44%). Both parents and providers believed that HPV vaccination provided important health benefits, but the timing of vaccination with relation to sexual activity was an important theme related to vaccine delay. Providers with lower self-reported vaccination rates delayed vaccine recommendations in girls perceived to be at low risk for sexual activity, and several parents reported that their providers suggested or supported delaying vaccination until their daughters were older. However, parents/guardians and providers agreed that predicting the timing of sexual debut was extremely difficult. In contrast, providers with high vaccination rates presented HPV vaccination as a routine vaccine with proven safety to prevent cancer, and parents responded positively to these messages. CONCLUSIONS:Although most parents and providers believe that HPV vaccination is important, missed opportunities result from assumptions about the timing of vaccination relative to sexual activity. Routinely recommending HPV vaccination as cancer prevention to be coadministered with other vaccines at age 11 years can improve vaccination rates. Pediatrics 2014;134:e666-e674 AUTHORS:

The use of functional levels to stratify treatment-related outcomes by pretreatment functional status and to display the proportions of patients with improved, stable, or worsened function after treatment provides information that more specifically conveys the expected impact of treatment to patients choosing among localized prostate cancer treatments.