Large language models (LLMs) that have been trained on multilingual but not parallel text exhibit a remarkable ability to translate between languages. We probe this ability in an in-depth study of the pathways language model (PaLM), which has demonstrated the strongest machine translation (MT) performance among similarly-trained LLMs to date. We investigate various strategies for choosing translation examples for few-shot prompting, concluding that example quality is the most important factor. Using optimized prompts, we revisit previous assessments of PaLM's MT capabilities with more recent test sets, modern MT metrics, and human evaluation, and find that its performance, while impressive, still lags that of state-of-the-art supervised systems. We conclude by providing an analysis of PaLM's MT output which reveals some interesting properties and prospects for future work.

Model-based reinforcement learning (RL) often achieves higher sample efficiency in practice than model-free RL by learning a dynamics model to generate samples for policy learning. Previous works learn a dynamics model that fits under the empirical state-action visitation distribution for all historical policies, i.e., the sample replay buffer. However, in this paper, we observe that fitting the dynamics model under the distribution for \emph{all historical policies} does not necessarily benefit model prediction for the \emph{current policy} since the policy in use is constantly evolving over time. The evolving policy during training will cause state-action visitation distribution shifts. We theoretically analyze how this distribution shift over historical policies affects the model learning and model rollouts. We then propose a novel dynamics model learning method, named \textit{Policy-adapted Dynamics Model Learning (PDML)}. PDML dynamically adjusts the historical policy mixture distribution to ensure the learned model can continually adapt to the state-action visitation distribution of the evolving policy. Experiments on a range of continuous control environments in MuJoCo show that PDML achieves significant improvement in sample efficiency and higher asymptotic performance combined with the state-of-the-art model-based RL methods.

Empirical studies suggest a deep intertwining between opinion formation and decision-making processes, but these have been treated as separate problems in the study of dynamical models for social networks. In this paper, we bridge the gap in the literature by proposing a novel coevolutionary model, in which each individual selects an action from a binary set and has an opinion on which action they prefer. Actions and opinions coevolve on a two-layer network. For homogeneous parameters, undirected networks, and under reasonable assumptions on the asynchronous updating mechanics, we prove that the coevolutionary dynamics is an ordinal potential game, enabling analysis via potential game theory. Specifically, we establish global convergence to the Nash equilibria of the game, proving that actions converge in a finite number of time steps, while opinions converge asymptotically. Next, we provide sufficient conditions for the existence of, and convergence to, polarized equilibria, whereby the population splits into two communities, each selecting and supporting one of the actions. Finally, we use simulations to examine the social psychological phenomenon of pluralistic ignorance.

With a low electrical hum, a small team of boxy, wheeled robots called "ants" criss-cross the top of a giant 3D grid of grey storage crates – 60,000 of them - ceaselessly arranging and rearranging them to order. Just one man, jokingly known as the robot whisperer, walks among them with a laptop. It would be hard to conceive of a more vivid example of robots taking on human jobs. "As robot technology advances, we can use them more and more, together with humans, to do useful work, and I think this is the future," says Jeroen Dekker, co-founder of Active Ants, the Dutch firm behind this newly opened e-commerce warehouse outside Northampton. "Yes, some jobs are disappearing, but that's the nasty jobs, for which we cannot find enough people."

A novelist responds to Jeff Hewitt's "The Big Four v. ORWELL." For the first time in history, a machine is capable of crafting flash fiction stories, poems, parody Bible verses, and spoof My Little Pony episode summaries, to everyone's delight (or horror). Narrative art, once thought the sole province of humans, has been invaded by large language models. Hollywood writers have told me they're terrified that studios will fire them all and fill writers' rooms with robots in a few years. Before we've even had a chance to absorb the fact that the Turing test (used to determine if an artificial intelligence can pass as human) has been demolished, it seems we writers are being handed pink slips.

Now Nasa engineers say they are developing their own ChatGPT-style interface that could ultimately allow astronauts to talk to their spacecraft and mission controllers to converse with artificial intelligence-powered robots exploring distant planets and moons. An early incarnation of the AI is slated to be deployed on Lunar Gateway, a planned extraterrestrial space station that is part of the Artemis programme, according to the engineer developing the technology. "The idea is to get to a point where we have conversational interactions with space vehicles and they [are] also talking back to us on alerts, interesting findings they see in the solar system and beyond," Dr Larissa Suzuki, a visiting researcher at Nasa said. Speaking at a meeting on next-generation space communication at the Institute of Electrical and Electronics Engineers (IEEE) in London on Tuesday, Suzuki outlined an interplanetary communications network with inbuilt AI to detect, and possibly fix, glitches and inefficiencies as they occur. "It then alerts mission operators that there is a likelihood that package transmissions from space vehicle X will be lost or will fail delivery," she said.

The New Zealand government identified numerous challenges related to trust and truth in the context of digital technologies. These challenges result from an ever-increasing amount of online social networks, end-to-end digital supply chains, automated decision-making tools, generative artificial intelligence (AI), and cyber-physical systems. Such challenges impact people's lives across professional and private contexts and led to the Veracity Projecta 2021–2024. Outside the field of computing, veracity is not a common term in everyday language. One dictionary definition is "conformity with truth or fact."b

Physics-informed machine learning (PIML) is a set of methods and tools that systematically integrate machine learning (ML) algorithms with physical constraints and abstract mathematical models developed in scientific and engineering domains. As opposed to purely data-driven methods, PIML models can be trained from additional information obtained by enforcing physical laws such as energy and mass conservation. More broadly, PIML models can include abstract properties and conditions such as stability, convexity, or invariance. The basic premise of PIML is that the integration of ML and physics can yield more effective, physically consistent, and data-efficient models. This paper aims to provide a tutorial-like overview of the recent advances in PIML for dynamical system modeling and control. Specifically, the paper covers an overview of the theory, fundamental concepts and methods, tools, and applications on topics of: 1) physics-informed learning for system identification; 2) physics-informed learning for control; 3) analysis and verification of PIML models; and 4) physics-informed digital twins. The paper is concluded with a perspective on open challenges and future research opportunities.

Randomness supports many critical functions in the field of machine learning (ML) including optimisation, data selection, privacy, and security. ML systems outsource the task of generating or harvesting randomness to the compiler, the cloud service provider or elsewhere in the toolchain. Yet there is a long history of attackers exploiting poor randomness, or even creating it -- as when the NSA put backdoors in random number generators to break cryptography. In this paper we consider whether attackers can compromise an ML system using only the randomness on which they commonly rely. We focus our effort on Randomised Smoothing, a popular approach to train certifiably robust models, and to certify specific input datapoints of an arbitrary model. We choose Randomised Smoothing since it is used for both security and safety -- to counteract adversarial examples and quantify uncertainty respectively. Under the hood, it relies on sampling Gaussian noise to explore the volume around a data point to certify that a model is not vulnerable to adversarial examples. We demonstrate an entirely novel attack against it, where an attacker backdoors the supplied randomness to falsely certify either an overestimate or an underestimate of robustness. We demonstrate that such attacks are possible, that they require very small changes to randomness to succeed, and that they can be hard to detect. As an example, we hide an attack in the random number generator and show that the randomness tests suggested by NIST fail to detect it. We advocate updating the NIST guidelines on random number testing to make them more appropriate for safety-critical and security-critical machine-learning applications.

Drug repurposing (or repositioning) is the process of finding new therapeutic uses for drugs already approved by drug regulatory authorities (e.g., the Food and Drug Administration (FDA) and Therapeutic Goods Administration (TGA)) for other diseases. This involves analyzing the interactions between different biological entities, such as drug targets (genes/proteins and biological pathways) and drug properties, to discover novel drug-target or drug-disease relations. Artificial intelligence methods such as machine learning and deep learning have successfully analyzed complex heterogeneous data in the biomedical domain and have also been used for drug repurposing. This study presents a novel unsupervised machine learning framework that utilizes a graph-based autoencoder for multi-feature type clustering on heterogeneous drug data. The dataset consists of 438 drugs, of which 224 are under clinical trials for COVID-19 (category A). The rest are systematically filtered to ensure the safety and efficacy of the treatment (category B). The framework solely relies on reported drug data, including its pharmacological properties, chemical/physical properties, interaction with the host, and efficacy in different publicly available COVID-19 assays. Our machine-learning framework reveals three clusters of interest and provides recommendations featuring the top 15 drugs for COVID-19 drug repurposing, which were shortlisted based on the predicted clusters that were dominated by category A drugs. The anti-COVID efficacy of the drugs should be verified by experimental studies. Our framework can be extended to support other datasets and drug repurposing studies, given open-source code and data availability.