Search for: All records

With the rapid improvement of large language models capabilities, there has been increasing interest in challenging constrained text generation problems. However, existing benchmarks for constrained generation usually focus on fixed constraint types (e.g. generate a sentence containing certain words) that have proved to be easy for state-of-the-art models like GPT-4. We present COLLIE, a grammar- based framework that allows the specification of rich, compositional constraints with diverse generation levels (word, sentence, paragraph, passage) and modeling challenges (e.g. language understanding, logical reasoning, counting, semantic planning). We also develop tools for automatic extraction of task instances given a constraint structure and a raw text corpus. Using COLLIE, we compile the COLLIE- v1 dataset with 2,080 instances comprising 13 constraint structures. We perform systematic experiments across five state-of-the-art instruction-tuned language mod- els and analyze their performances to reveal shortcomings. COLLIE is designed to be extensible and lightweight, and we hope the community finds it useful to develop more complex constraints and evaluations in the future. 

We demonstrate a simple and compact variable magnetic field source based on the permanent cube magnet array approximating a Halbach cylinder. The large air gap area accommodates standard cryostat tails while providing a high uniformity and magnetic field stability of up to 0.5 T over regions of up to about a centimeter. It eliminates magnetic remanence effects and produces reproducible fields without the need for feedback. Thanks to the low cost and exceptional energy efficiency, it provides an accessible solution for modest magnetic field requirements in a wide range of research applications.

 

Cybersecurity and Artificial Intelligence (AI) are key domains whose intersection gives great promises and poses significant threats. Indeed, the National Academy of Science (NAS), the National Science Foundation (NSF), and othßer respected entities have noted the significant role that AI can play in cybersecurity, and the importance of ensuring the security of AI-enabled algorithms and systems. This minitrack focuses on AI and Cybersecurity that works in broader domains, collaborative inter-organizational realms, shared collaborative domains, or with collaborative technologies. The papers in this minitrack have the potential to offer interesting and impactful solutions to emerging areas, including unmanned aerial vehicles and open source software security. 

Site U1595 (proposed Site CSK-08B) is located in the southern basin of the Santorini caldera at 291 meters below sea level (mbsl) (Figure F1). It was drilled in three holes (U1595A–U1595C) to a maximum recovery depth of 127.0 meters below seafloor (mbsf) (all depths below seafloor are given using the core depth below seafloor, Method A [CSF-A], scale, except in Operations where the drilling depth below seafloor [DSF] scale is used). Average hole core recoveries range 46%–77%. Site U1594 addresses the same drilling objectives and lies northeast of Site U1595. Two additional sites (U1596 and U1597) lie in the northern caldera basin. Four seismic units (S1–S4) were previously recognized in the Santorini caldera (Johnston et al., 2015; Nomikou et al., 2016) (Figure F2). Seismic Units S1–S3 were thought to consist of muds and sands from cliff mass wasting (Unit S1); compacted (possibly lithified) sandy volcaniclastics from Kameni Volcano (Unit S2); and consolidated coarse blocky intracaldera tuffs, landslide debris, and/or flood gravels (Unit S3). Unit S4 was thought to be intracaldera tuff from the Late Bronze Age eruption. The four caldera sites were planned to sample Units S1–S3; test the published correlations between the two caldera basins; penetrate below Unit S3 into Unit S4; and address scientific Objectives 1, 4, 5, and 7 of the Expedition 398 Scientific Prospectus (Druitt et al., 2022). By drilling both caldera basins and exploiting our dense seismic reflection coverage, we gained access to the 3D architecture of the entire caldera fill. We also targeted the question of why the northern basin is 100 m deeper than the southern one. Finally, we tested whether Unit S3 consisted of flood debris from the caldera flooding event (Nomikou et al., 2016) or whether it was Late Bronze Age intracaldera tuff (Johnston et al., 2015). The intracaldera sites were used for microbiological work of Objective 7.