While existing alignment paradigms have been integral in developing large language models (LLMs), LLMs often learn an averaged human preference and struggle to model diverse preferences across cultures, demographics, and communities. We propose Modular Pluralism, a modular framework based on multi-LLM collaboration for pluralistic alignment: it "plugs into" a base LLM a pool of smaller but specialized community LMs, where models collaborate in distinct modes to flexibility support three modes of pluralism: Overton, steerable, and distributional. Modular Pluralism is uniquely compatible with black-box LLMs and offers the modular control of adding new community LMs for previously underrepresented communities. We evaluate Modular Pluralism with six tasks and four datasets featuring questions/instructions with value-laden and perspective-informed responses. Extensive experiments demonstrate that Modular Pluralism advances the three pluralism objectives across six black-box and open-source LLMs. Further analysis reveals that LLMs are generally faithful to the inputs from smaller community LLMs, allowing seamless patching by adding a new community LM to better cover previously underrepresented communities.

Modern language models, while sophisticated, exhibit some inherent shortcomings, particularly in conversational settings. We claim that many of the observed shortcomings can be attributed to violation of one or more conversational principles. By drawing upon extensive research from both the social science and AI communities, we propose a set of maxims -- quantity, quality, relevance, manner, benevolence, and transparency -- for describing effective human-AI conversation. We first justify the applicability of the first four maxims (from Grice) in the context of human-AI interactions. We then argue that two new maxims, benevolence (concerning the generation of, and engagement with, harmful content) and transparency (concerning recognition of one's knowledge boundaries, operational constraints, and intents), are necessary for addressing behavior unique to modern human-AI interactions. We evaluate the degree to which various language models are able to understand these maxims and find that models possess an internal prioritization of principles that can significantly impact their ability to interpret the maxims accurately.

In this paper, we tackle a new problem of \textit{multi-source unsupervised domain adaptation (MSUDA) for graphs}, where models trained on annotated source domains need to be transferred to the unsupervised target graph for node classification. Due to the discrepancy in distribution across domains, the key challenge is how to select good source instances and how to adapt the model. Diverse graph structures further complicate this problem, rendering previous MSUDA approaches less effective. In this work, we present the framework Selective Multi-source Adaptation for Graph ({\method}), with a graph-modeling-based domain selector, a sub-graph node selector, and a bi-level alignment objective for the adaptation. Concretely, to facilitate the identification of informative source data, the similarity across graphs is disentangled and measured with the transferability of a graph-modeling task set, and we use it as evidence for source domain selection. A node selector is further incorporated to capture the variation in transferability of nodes within the same source domain. To learn invariant features for adaptation, we align the target domain to selected source data both at the embedding space by minimizing the optimal transport distance and at the classification level by distilling the label function. Modules are explicitly learned to select informative source data and conduct the alignment in virtual training splits with a meta-learning strategy. Experimental results on five graph datasets show the effectiveness of the proposed method.

The United States Department of the Treasury has fleshed out a proposed rule that would restrict and monitor US investments in China for artificial intelligence, computer chips and quantum computing. The fleshed-out draft rule, issued on Friday, stems from President Joe Biden's August executive order regarding the access that "countries of concern" have to American dollars to fund advanced technologies that could enhance those nations' military, intelligence, surveillance and cyber-capabilities. The order identified China, Hong Kong and Macau as countries of concern. The Biden administration has sought to stymie the development of technologies by China, the world's second largest economy, that could give it a military edge or enable it to dominate emerging sectors such as electric vehicles (EVs). In addition to the proposed rule, Biden, a Democrat, has also placed a stiff tariff on Chinese EVs, an issue with political implications as Biden and his Republican presidential opponent Donald Trump are both trying to show voters who can best stand up to China, a geopolitical rival and major trading partner.

Apple will delay launching three new artificial intelligence features in Europe because European Union competition rules require the company ensure that rival products and services can function with its devices. The features will launch in the fall in the US but will not arrive in Europe until 2025. The company said on Friday three features – Phone Mirroring, SharePlay Screen Sharing enhancements, and Apple Intelligence – will not be rolled out to EU users this year because of regulatory uncertainties due to the EU's Digital Markets Act (DMA). Apple said the EU's regulations would force it to compromise its devices' security, an argument it has made before and that EU officials have pushed back on. "Specifically, we are concerned that the interoperability requirements of the DMA could force us to compromise the integrity of our products in ways that risk user privacy and data security," Apple said in an email.

Samuel Bendett says Russia aims to dominate in AI development and implementation, but faces a number of hurdles that could force officials to pursue risky alternatives to cover the gaps in development. Russia sees a pressing need for security talks with the United States but they must be "comprehensive" and include the subject of Ukraine, the Kremlin said on Friday. "It is impossible to rip out any individual segments from the general complex of accumulated problems, and we will not do this," Kremlin spokesman Dmitry Peskov said when asked if Moscow was ready to talk to Washington about nuclear risks. "So we are open to dialogue, but to a broad comprehensive dialogue that covers all dimensions, including the current dimension related to the conflict around Ukraine, related to the direct involvement of the USA in this conflict," Peskov told reporters. The United States rejects Russia's contention that by arming Ukraine it has become a direct protagonist in a war aimed at inflicting a crushing "strategic defeat" on Moscow.

A few weeks ago, I witnessed Google Search make what could have been the most expensive error in its history. In response to a query about cheating in chess, Google's new AI Overview told me that the young American player Hans Niemann had "admitted to using an engine," or a chess-playing AI, after defeating Magnus Carlsen in 2022--implying that Niemann had confessed to cheating against the world's top-ranked player. Suspicion about the American's play against Carlsen that September indeed sparked controversy, one that reverberated even beyond the world of professional chess, garnering mainstream news coverage and the attention of Elon Musk. Except, Niemann admitted no such thing. Quite the opposite: He has vigorously defended himself against the allegations, going so far as to file a 100 million defamation lawsuit against Carlsen and several others who had accused him of cheating or punished him for the unproven allegation--Chess.com, for example, had banned Niemann from its website and tournaments.

His lab uses mice for some protocols, but animal studies are notoriously bad at identifying human treatments. Around 95% of the drugs developed through animal research fail in people. Researchers have documented this translation gap since at least 1962. "All these pharmaceutical companies know the animal models stink," says Don Ingber, founder of the Wyss Institute for Biologically Inspired Engineering at Harvard and a leading advocate for organs on chips. "The FDA knows they stink."

Russia increasingly looks toward artificial intelligence (AI) to address deficiencies in its battlefield capabilities and capacities that the invasion of Ukraine has exposed, according to experts. "Russian futurists, Russian technologists, Russian developers are envisioning this slow evolution away from larger human involvement to where humans are going to be involved as little as possible," Samuel Bendett, adjunct senior fellow in the Technology and National Security Program at the Center for a New American Security (CNAS), told Fox News Digital. "Some of those statements were made prior to Russia's disastrous invasion of Ukraine and Russia's conduct in this war, which is very much manpower intensive… but this is something that the Russian military is keeping sort of on the horizon," he said. Bendett in his paper for CNAS argued that Russia's keenness to adopt AI could lead the country to take greater risks as it seeks to catch up with the West. He relied on public statements, announcements and analysis of Russian-language media to develop his paper, which looks at major developments in robotics and AI spaces and as Russia seeks an "intellectualized" military that makes semiautomated decisions.

Projecting sea-level change in various climate-change scenarios typically involves running forward simulations of the Earth's gravitational, rotational and deformational (GRD) response to ice mass change, which requires high computational cost and time. Here we build neural-network emulators of sea-level change at 27 coastal locations, due to the GRD effects associated with future Antarctic Ice Sheet mass change over the 21st century. The emulators are based on datasets produced using a numerical solver for the static sea-level equation and published ISMIP6-2100 ice-sheet model simulations referenced in the IPCC AR6 report. We show that the neural-network emulators have an accuracy that is competitive with baseline machine learning emulators. In order to quantify uncertainty, we derive well-calibrated prediction intervals for simulated sea-level change via a linear regression postprocessing technique that uses (nonlinear) machine learning model outputs, a technique that has previously been applied to numerical climate models. We also demonstrate substantial gains in computational efficiency: a feedforward neural-network emulator exhibits on the order of 100 times speedup in comparison to the numerical sea-level equation solver that is used for training.