Search for: All records

The development of generative language models that can create long and coherent textual outputs via autoregression has lead to a proliferation of uses and a corresponding sweep of analyses as researches work to determine the limitations of this new paradigm. Unlike humans, these ‘Large Language Models’ (LLMs) are highly sensitive to small changes in their inputs, leading to unwanted inconsistency in their behavior. One problematic inconsistency when LLMs are used to answer multiple-choice questions or analyze multiple inputs is order dependency: the output of an LLM can (and often does) change significantly when sub-sequences are swapped, despite both orderings being semantically identical. In this paper we present Set-Based Prompting, a technique that guarantees the output of an LLM will not have order dependence on a specified set of sub-sequences. We show that this method provably eliminates order dependency, and that it can be applied to any transformer-based LLM to enable text generation that is unaffected by re-orderings. Delving into the implications of our method, we show that, despite our inputs being out of distribution, the impact on expected accuracy is small, where the expectation is over the order of uniformly chosen shuffling of the candidate responses, and usually significantly less in practice. Thus, Set-Based Prompting can be used as a ‘dropped-in’ method on fully trained models. Finally, we discuss how our method’s success suggests that other strong guarantees can be obtained on LLM performance via modifying the input representations. Code is available at github.com/reidmcy/set-based-prompting. 

Narrative planning is the use of automated planning to construct, communicate, and understand stories, a form of information to which human cognition and enaction is pre-disposed. We review the narrative planning problem in a manner suitable as an introduction to the area, survey different plan-based methodologies and affordances for reasoning about narrative, and discuss open challenges relevant to the broader AI community. 

Abstract Anaerobic gut fungi (AGF; Neocallimastigomycota) are crucial for the degradation of plant biomass in herbivores. While extensively studied in mammals, information regarding their occurrence, diversity, and community structure in nonmammalian hosts remains sparse. Here, we report on the AGF community in fecal samples of 13 domesticated ostriches. The ostrich (Struthio camelus) is an herbivorous, flightless, hindgut-fermenting member of the class Aves (birds). Illumina-based metabarcoding targeting the D2 region of the large ribosomal subunit (28S rRNA) revealed a uniform AGF community with low alpha diversity in the fecal samples. The community was mostly comprised of sequences potentially representing two novel species in the genus Piromyces, and a novel genus in the Neocallimastigomycota. Sequences affiliated with these novel taxa were absent or extremely rare in datasets derived from mammalian and tortoise samples, indicating a strong pattern of AGF-host association. One Piromyces strain (strain Ost1) was successfully isolated. Transcriptomics-enabled molecular dating analysis suggested a divergence time of ≈ 30Mya, a time frame in line with current estimates for ostrich evolution. Comparative gene content analysis between strain Ost1 and other Piromyces species from mammalian sources revealed a high degree of similarity. Our findings expand the range of AGF animal hosts to include members of the birds (class Aves), highlight a unique AGF community in the ostrich alimentary tract, and document the occurrence of a strong pattern of fungal–host association in ostriches, similar to previously observed patterns in AGF canonical mammalian hosts. 

Hurricanes have been the most expensive and devastating natural disasters in US history. Nevertheless, our understanding of the turbulence processes in these flows is limited due to the lack of sufficient measurements and high-resolution turbulence-resolving simulations. Our objective in this study is to bridge this knowledge gap by conducting a thorough sensitivity analysis of the hurricane boundary layer using large-eddy simulations. We characterize the impacts of the radius, surface roughness, gradient wind, and sub-grid scale viscosity models on the hurricane’s mean and turbulence dynamics. Our results indicate that increasing the Rossby number (rotation) increases the hurricane’s maximum jet velocity, reduces the boundary layer height, and decreases the size of coherent turbulent structures at the same elevation. This study provides insights into the turbulence dynamics in hurricanes and can guide the development of more accurate turbulence models for rotating hurricane flows. 

This paper describes how intentional and unintentional radio emission from airplanes is recorded with the Radio Neutrino Observatory Greenland (RNO-G). We characterize the received signals and define a procedure to extract a clean set of impulsive signals. These signals are highly suitable for instrument calibration, also for future experiments. A set of signals is used to probe the timing precision of RNO-G in-situ, which is found to match expectations. We also discuss the impact of these signals on the ability to detect neutrinos with RNO-G. 

The Radio Neutrino Observatory in Greenland (RNO-G) is the first in-ice radio array in the northern hemisphere for the detection of ultra-high energy neutrinos via the coherent radio emission from neutrino-induced particle cascades within the ice. The array is currently in phased construction near Summit Station on the Greenland ice sheet, with 7 stations deployed during the first two boreal summer field seasons of 2021 and 2022. In this paper, we describe the installation and system design of these initial RNO-G stations, and discuss the performance of the array as of summer 2024.