Search

Large Language Models (LLMs) have demonstrated impressive abilities in recent years with regards to code generation and understanding. However, little work has investigated how documentation and other code properties affect an LLM's ability to understand and generate code or documentation. We present an empirical analysis of how underlying properties of code or documentation can affect an LLM's capabilities. We show that providing an LLM with "incorrect" documentation can greatly hinder code understanding, while incomplete or missing documentation does not seem to significantly affect an LLM's ability to understand code., Comment: 7 pages, 5 figures, 2 tables. Accepted as a Findings paper in the "Generation" track to NAACL 2024. MITRE Public Release Case Number 23-4132

Neural networks have shown great potential in compressing volume data for visualization. However, due to the high cost of training and inference, such volumetric neural representations have thus far only been applied to offline data processing and non-interactive rendering. In this paper, we demonstrate that by simultaneously leveraging modern GPU tensor cores, a native CUDA neural network framework, and a well-designed rendering algorithm with macro-cell acceleration, we can interactively ray trace volumetric neural representations (10-60fps). Our neural representations are also high-fidelity (PSNR > 30dB) and compact (10-1000x smaller). Additionally, we show that it is possible to fit the entire training step inside a rendering loop and skip the pre-training process completely. To support extreme-scale volume data, we also develop an efficient out-of-core training strategy, which allows our volumetric neural representation training to potentially scale up to terascale using only an NVIDIA RTX 3090 workstation., Comment: There is a supplementary video for this manuscript, which can be accessed via this link: https://drive.google.com/file/d/17wSgIm_VsoeGhfyZwMpOnCYy2Mj3ydGv/view?usp=sharing

Responses to a survey of the environment for research in the natural sciences at predominantly undergraduate universities were received from 136 institutional representatives and more than 2,900 faculty members. There was a nearly uniform acknowledgement that time was a limiting factor for research, and that faculty had reached their limits. Faculty at undergraduate institutions have seen a reduction in their formal teaching loads over the last 40 years, as data from other studies show, but the increased emphasis on active learning and discovery-based education and increased expectations of research engagements with students are taking greater amounts of faculty time. Faculty make additional efforts to attract and retain students and spend increasing amounts of time on administrative duties without being compensated by release time from other duties. Both teaching and research require more time than they did in the past. Time is not the only factor limiting research, but it appears to be an important one. (SLD)

One of the strongest arguments made to attract increased support for science for undergraduate liberal arts colleges in the mid- to late-1980s was that the proportion of students obtaining their Ph.D. degrees was larger at these institutions than at their Ph.D. granting university counterparts. This conclusion was supported by later research, but whether it remains meaningful today was one subject of research by the Research Corporation. The recent data compilation for the natural sciences at undergraduate institutions, published as "Academic Excellence: The Sourcebook," provided updated information on science degrees and the baccalaureate origins of Ph.D.s. These data make it evident that there is significant variation in the ratio of bachelor's degrees to Ph.D.s. Also noteworthy is that institutions with large enrollments and relatively large numbers of students can produce a greater number of students who obtain their Ph.D. degrees than many smaller institutions. When current findings are compared with those from the 1980s, it is evident that fully two-thirds of the institutions studied lost ground in terms of the ratio of bachelor's degrees to Ph.D.s. Some limitations of the data available are noted. (SLD)

Using grant and publication data from nearly 3,000 faculty members in the natural sciences at 133 predominantly undergraduate universities, this study investigated the characteristics of faculty who received $1,000,000 or more in external grant support during the 1990s. This selection required the faculty member to receive $100,000 annually through 10 years. As is consistent with Zipf's Law, (G. K. Zipf; the probability of an item starts high and then tapers off; a few occur very often, while others occur rarely) only about 51 of the 2,980 natural sciences faculty responding (1.7%) received these high grant amounts. The paper reports the relevant characteristics of rank, gender, and discipline for the high grant recipients. However, these million-dollar faculty do not provide a clear pattern of drawing students into science. A study of faculty members with 30 peer-reviewed publications, only 38 of 2,980, shows that these faculty also do not demonstrate a clear advantage in drawing students into science. Overall findings show that is the rare faculty member who chooses to conduct a significant research program that involves undergraduate students in meaningful ways. (SLD)

We have designed, constructed and put into operation a very large area CCD camera that covers the field of view of the 1.2 m Samuel Oschin Schmidt Telescope at the Palomar Observatory. The camera consists of 112 CCDs arranged in a mosaic of four rows with 28 CCDs each. The CCDs are 600 x 2400 pixel Sarnoff thinned, back illuminated devices with 13 um x 13 um pixels. The camera covers an area of 4.6 deg x 3.6 deg on the sky with an active area of 9.6 square degrees. This camera has been installed at the prime focus of the telescope, commissioned, and scientific quality observations on the Palomar-QUEST Variability Sky Survey were started in September of 2003. The design considerations, construction features, and performance parameters of this camera are described in this paper.

We have converted the Palomar 60-inch telescope (P60) from a classical night assistant-operated telescope to a fully robotic facility. The automated system, which has been operational since September 2004, is designed for moderately fast (t <~ 3 minutes) and sustained (R <~ 23 mag) observations of gamma-ray burst afterglows and other transient events. Routine queue-scheduled observations can be interrupted in response to electronic notification of transient events. An automated pipeline reduces data in real-time, which is then stored on a searchable web-based archive for ease of distribution. We describe here the design requirements, hardware and software upgrades, and lessons learned from roboticization. We present an overview of the current system performance as well as plans for future upgrades., Comment: Accepted in PASP; 26 pages, 7 figures; high resolution version at http://www.srl.caltech.edu/~cenko/public/papers/p60.ps

The decade of the 1980s saw a shift in paradigm in American trade and commerce, the serious beginning of a movement away from anachronistic and unwise ethnocentrism and nationalism and toward a more practical globalization of American business consciousness. The notion of business conducted in English and according to American norms began to yield to internationalization. Higher education responded by creating interdisciplinary programs combining study of business with study of languages and cultures of potential business partners and co-workers. Within the new educational epistemology created by this movement, Spanish is a particularly important element, as a major world language, the dominant language of the hemisphere of the Americas, and an emerging major language in the United States. Business Spanish will and must play an increasingly important role in preparing graduates for the national, hemispheric, and global challenges and opportunities ahead. (MSE)

The situation card used in the American Council on the Teaching of Foreign Languages/Educational Testing Service (ACTFL/ETS) Oral Proficiency Interview provides a kind of linguistic-topographical map of the student's foreign language ability by measuring what the student of business Spanish is able or unable to do in business situations in that language. The situation card can reveal the particulars of grammar, vocabulary, and fluency on which the student still needs work. A pool of business Spanish situation cards can be adapted to reflect and enhance any aspect of course content and reinforce target vocabulary. The cards also provide spontaneity that develops a realistic student flexibility. The situations seem to be most effective when they are presented in English so that students do not misunderstand the directions and so that key vocabulary words and phrases are not revealed. Cards can be used regularly, perhaps at the end of each class session. Individual performances should be tape-recorded for student review. A selection of situation cards adapted from the ACTFL/ETS model are presented, and the procedure for their use is reviewed. The situations are designed for three proficiency levels (intermediate, advanced, and superior) and contain a variety of scenario types and topics. (MSE)