Random Utility Models (RUMs) are a classical framework for modeling user preferences and play a key role in reward modeling for Reinforcement Learning from Human Feedback (RLHF). However, a crucial shortcoming of many of these techniques is the Independence of Irrelevant Alternatives (IIA) assumption, which collapses \emph{all} human preferences to a universal underlying utility function, yielding a coarse approximation of the range of human preferences. On the other hand, statistical and computational guarantees for models avoiding this assumption are scarce. In this paper, we investigate the statistical and computational challenges of learning a \emph{correlated} probit model, a fundamental RUM that avoids the IIA assumption. First, we establish that the classical data collection paradigm of pairwise preference data is \emph{fundamentally insufficient} to learn correlational information, explaining the lack of statistical and computational guarantees in this setting. Next, we demonstrate that \emph{best-of-three} preference data provably overcomes these shortcomings, and devise a statistically and computationally efficient estimator with near-optimal performance. These results highlight the benefits of higher-order preference data in learning correlated utilities, allowing for more fine-grained modeling of human preferences. Finally, we validate these theoretical guarantees on several real-world datasets, demonstrating improved personalization of human preferences.

When a sophisticated Chinese spy balloon floated over America in early 2023, lawmakers and the public were outraged at the Biden-Harris administration's passivity and initial inclination to keep it quiet – only acknowledging the balloon after two civilian photographers forced their hand. Now, the Wall Street Journal has broken news on an even more stupendous U.S. national security breach, reporting that drones flew over a sensitive nuclear weapons testing facility for three days last October and then, two months later, flew over Langley Air Force Base in Virginia for 17 straight nights while the Biden White House, and the military officers it promoted, dawdled and argued over what to do about it. The swarms started on Dec. 7, 2023. Drones, some as large as 20 feet long, flew at night over the Air Combat Command headquarters with its squadrons of advanced F-22 Raptor fighters. The blame-casting and responsibility-shirking reveal a dangerous pattern of hesitation and risk-averse decision-making.

As with most every other aspect of modern society, computerization, augmentation and automation have hyper-accelerated the pace at which wars are prosecuted -- and who better to help reshape the US military into a 21st century fighting force than an entire industry centered on moving fast and breaking things? In his latest book, War Virtually: The Quest to Automate Conflict, Militarize Data, and Predict the Future, professor and chair of the Anthropology Department at San José State University, Roberto J González examines the military's increasing reliance on remote weaponry and robotic systems are changing the way wars are waged. In the excerpt below, González investigates Big Tech's role in the Pentagon's high-tech transformations. Excerpted from War Virtually: The Quest to Automate Conflict, Militarize Data, and Predict the Future by Roberto J. González, published by the University of California Press. Ash Carter's plan was simple but ambitious: to harness the best and brightest ideas from the tech industry for Pentagon use.

Playing a mob boss in a video game might not seem all that serious for a veteran actor who has appeared on feature films and television. For George Takei, however, performing the role of Masumi Arakawa in "Yakuza: Like a Dragon," is akin to his life coming full circle. Takei was only six years old when he was first introduced to the concept of "benshi" -- Japanese performers who narrated for silent films. It was a discovery that happened to coincide with a dark period in American history. At the time, Takei and his family were staying in Arkansas, part of some 120,000 Japanese-Americans who were gathered into internment camps after Pearl Harbor was bombed on Dec. 7, 1941.

The National Oceanic and Atmospheric Administration has released photographs of a Japanese mini-submarine that was sunk at the very beginning of the Pearl Harbor attack in 1941, and they're haunting. On Wednesday, the 75th anniversary of the "date which will live in infamy," the NOAA ship Okeanos Explorer dispatched a robotic vehicle to explore two Japanese mini submarines, an event that they live-streamed. In what marks the first US shots fired in World War II, the USS Ward fired at one mini-sub on the morning of December 7, 1941, sinking it, after it was first spotted attempting to enter the harbor, partially submerged. Ninety minutes after the Ward sunk the sub, the aerial attack by the Japanese began. The first submarine NOAA explored-- the one sunk by the Ward-- had changed a great deal since the last visit to the wreck, with a peeling hull, two sections that separated from one another, and lots of biological growth, NOAA scientists report.