On an airstrip somewhere in Texas, a swarm of killer jets approaches--controlled by, of all things, a large language model. At a secret US military base located about 50 miles from the Mexican border--exact location: classified--the defense contractor Anduril is testing a remarkable new use for a large language model. I attended one of the first demonstrations last year. From a sun-bleached landing strip, I watched as four jet aircraft, codenamed Mustang, appeared on the horizon to the west and soared over a desolate landscape of boulders and brush. The prototypes, miniaturized for the demo, fell into formation, their engines buzzing as they grew near.

Some of the world's most interesting thinkers about thinking think they might've cracked machine sentience. And I think they might be onto something. As a journalist who covers AI, I hear from countless people who seem utterly convinced that ChatGPT, Claude, or some other chatbot has achieved "sentience." The Turing test was aced a while back, yes, but unlike rote intelligence, these things are not so easily pinned down. Large language models will claim to think for themselves, even describe inner torments or profess undying loves, but such statements don't imply interiority.

I sound Korean--because I am Korean. Can AI make me sound American? It all began, as these things often do, with an Instagram ad . "No one tells you this if you're an immigrant, but accent discrimination is a real thing," said a woman in the video. Her own accent is faintly Eastern European--so subtle it took me a few playbacks to notice.

I tried OpenAI's new Atlas browser but I still don't know what it's for My impression is that it is little more than cynicism masquerading as software. OpenAI ChatGPT Atlas introducing is being displayed on a mobile phone with the company's branding seen in the background, in this photo illustration. Taken in Brussels, Belgium, on 23 October 2025. OpenAI rolled out a new web browser last week called Atlas. It comes with ChatGPT built in, along with an agent, so that you can browse, get direct answers, and have automated tasks performed on your behalf all at the same time. I've spent the past several days tinkering with Atlas.

ChatGPT, Gemini, DeepSeek, and Grok are serving users propaganda from Russian-backed media when asked about the invasion of Ukraine, new research finds. OpenAI's ChatGPT, Google's Gemini, DeepSeek, and xAI's Grok are pushing Russian state propaganda from sanctioned entities--including citations from Russian state media, sites tied to Russian intelligence or pro-Kremlin narratives--when asked about the war against Ukraine, according to a new report. Researchers from the Institute of Strategic Dialogue (ISD) claim that Russian propaganda has targeted and exploited data voids --where searches for real-time data provide few results from legitimate sources--to promote false and misleading information. Almost one-fifth of responses to questions about Russia's war in Ukraine, across the four chatbots they tested, cited Russian state-attributed sources, the ISD research claims. "It raises questions regarding how chatbots should deal when referencing these sources, considering many of them are sanctioned in the EU," says Pablo Maristany de las Casas, an analyst at the ISD who led the research.

As future superhuman models become increasingly complex, accurately supervising their behavior may exceed human capabilities. Recent works have demonstrated that in such scenarios, weak models can effectively supervise strong models, a phenomenon known as weak-to-strong generalization. However, we find that naive weak-to-strong generalization fails under distribution shifts, often leading to worse performance of the strong model than its weak supervisors. To address this, we propose RAVEN, a robust weak-to-strong generalization framework that dynamically learns the optimal combinations of weak models in addition to parameters of the strong model. We demonstrate the effectiveness of RAVEN on image classification, text classification, and preference alignment tasks. RAVEN outperforms alternative baselines by over 30% on out-of-distribution tasks while matching or surpassing existing methods on in-distribution tasks. Moreover, our results show that RAVEN assigns higher weights to more accurate weak models, demonstrating its ability to automatically identify trustworthy supervision.

Fine-tuning has proven to be highly effective in adapting pre-trained models to perform better on new desired tasks with minimal data samples. Among the most widely used approaches are reparameterization methods, which update a target module by augmenting its frozen weight matrix with an additional trainable weight matrix. The most prominent example is Low Rank Adaption (LoRA), which gained significant attention in recent years. In this paper, we introduce a new class of reparameterization methods for transfer learning, designed to enhance the generalization ability of fine-tuned models. We establish the effectiveness of our approach in a high-dimensional binary classification setting using tools from Random Matrix Theory, and further validate our theoretical findings through more realistic experiments, such as fine-tuning LLMs.

Large Language Models (LLMs) enable a new form of digital experimentation where treatments combine human and model-generated content in increasingly sophisticated ways. The main methodological challenge in this setting is representing these high-dimensional treatments without losing their semantic meaning or rendering analysis intractable. Here, we address this problem by focusing on learning low-dimensional representations that capture the underlying structure of such treatments. These representations enable downstream applications such as guiding generative models to produce meaningful treatment variants and facilitating adaptive assignment in online experiments. We propose double kernel representation learning, which models the causal effect through the inner product of kernel-based representations of treatments and user covariates. We develop an alternating-minimization algorithm that learns these representations efficiently from data and provides convergence guarantees under a low-rank factor model. As an application of this framework, we introduce an adaptive design strategy for online experimentation and demonstrate the method's effectiveness through numerical experiments.

Reinforcement learning (RL) has become a predominant technique to align language models (LMs) with human preferences or promote outputs which are deemed to be desirable by a given reward function. Standard RL approaches optimize average reward, while methods explicitly focused on reducing the probability of undesired outputs typically come at a cost to average-case performance. To improve this tradeoff, we introduce RePULSe, a new training method that augments the standard RL loss with an additional loss that uses learned proposals to guide sampling low-reward outputs, and then reduces those outputs' probability. We run experiments demonstrating that RePULSe produces a better tradeoff of expected reward versus the probability of undesired outputs and is more adversarially robust, compared to standard RL alignment approaches and alternatives.

Designing optimal prompts and reasoning processes for large language models (LLMs) on domain-specific tasks is both necessary and challenging in real-world applications. Determining how to integrate domain knowledge, enhance reasoning efficiency, and even provide domain experts with refined knowledge integration hints are particularly crucial yet unresolved tasks. In this research, we propose Evolutionary Graph Optimization for Prompting (EGO-Prompt), an automated framework to designing better prompts, efficient reasoning processes and providing enhanced causal-informed process. EGO-Prompt begins with a general prompt and fault-tolerant initial Semantic Causal Graph (SCG) descriptions, constructed by human experts, which is then automatically refined and optimized to guide LLM reasoning. Recognizing that expert-defined SCGs may be partial or imperfect and that their optimal integration varies across LLMs, EGO-Prompt integrates a novel causal-guided textual gradient process in two steps: first, generating nearly deterministic reasoning guidance from the SCG for each instance, and second, adapting the LLM to effectively utilize the guidance alongside the original input. The iterative optimization algorithm further refines both the SCG and the reasoning mechanism using textual gradients with ground-truth. We tested the framework on real-world public health, transportation and human behavior tasks. EGO-Prompt achieves 7.32%-12.61% higher F1 than cutting-edge methods, and allows small models to reach the performence of larger models at under 20% of the original cost. It also outputs a refined, domain-specific SCG that improves interpretability.