The rise of large-scale pretrained models has made it feasible to generate predictive or synthetic features at low cost, raising the question of how to incorporate such surrogate predictions into downstream decision-making. We study this problem in the setting of online linear contextual bandits, where contexts may be complex, nonstationary, and only partially observed. In addition to bandit data, we assume access to an auxiliary dataset containing fully observed contexts--common in practice since such data are collected without adaptive interventions. We propose PULSE-UCB, an algorithm that leverages pretrained models trained on the auxiliary data to impute missing features during online decision-making. We establish regret guarantees that decompose into a standard bandit term plus an additional component reflecting pretrained model quality. In the i.i.d. context case with Hölder-smooth missing features, PULSE-UCB achieves near-optimal performance, supported by matching lower bounds. Our results quantify how uncertainty in predicted contexts affects decision quality and how much historical data is needed to improve downstream learning.

If you love AI, you'll love Ken Liu's new cyberpunk thriller In Ken Liu's All That We See or Seem, a once-famous hacker must find a missing dream-weaver. The latest novel by Ken Liu, All That We See or Seem, is the near-future story of the mysterious disappearance of a professional dream-weaver called Elli. It is being marketed as a cyberpunk thriller . Full disclosure: I don't generally seek out thrillers or cyberpunk books, so I may not be the target audience for this. But I was keen to read it because Liu has not one but two claims to fame: as well as being the author of a celebrated fantasy series called The Dandelion Dynasty, he is also the translator of the sensationally good Remembrance of Earth's Past trilogy by Cixin Liu .

Artificial intelligence is becoming a permanent element in the world of work, with Silicon Valley calling it the dawning of a new age. Many people are afraid of losing their job, but Germany is well-prepared. In the northern part of the U.S. state of Louisiana, right next to the prison on the outskirts of Shreveport, looms a gigantic building of concrete and steel. Welcome to the future," reads a colorful greeting painted on the wall at the entrance, right next to the obligatory American flag. It is 9:30 a.m., a busy time of day. Yet the halls and corridors of SHV1, as the building is referred to internally, are completely empty of people. A blueprint for the future," as the site manager calls it. The Seattle-based company operates the largest fleet of industrial robots in the world, more than a million of them, and many are outfitted with artificial intelligence, helping them to lift, sort, search, weigh and scan. Guided and directed completely by AI. Without the massive use of this technology," says Aaron Parness, a former NASA aerospace engineer who now heads up the retail giant's AI robotic department, we would be a different company." The article you are reading originally appeared in German in issue 41/2025 (October 2nd, 2025) of DER SPIEGEL. Amazon, though, also employs people. But their role is changing rapidly.

Scientists hope to prevent deaths from climate change, but heat and cold are more complicated than we thought. Libby Cowgill is an anthropologist at the University of Missouri who hopes to revamp the science of thermoregulation. Libby Cowgill, an anthropologist in a furry parka, has wheeled me and my cot into a metal-walled room set to 40 F. A loud fan pummels me from above and siphons the dregs of my body heat through the cot's mesh from below. A large respirator fits snug over my nose and mouth. The device tracks carbon dioxide in my exhales--a proxy for how my metabolism speeds up or slows down throughout the experiment. Eventually Cowgill will remove my respirator to slip a wire-thin metal temperature probe several pointy inches into my nose. Cowgill and a graduate student quietly observe me from the corner of their so-called "climate chamber. Just a few hours earlier I'd sat beside them to observe as another volunteer, a 24-year-old personal trainer, endured the cold. Every few minutes, they measured his skin temperature with a thermal camera, his core temperature with a wireless pill, and his blood pressure and other metrics that hinted at how his body handles extreme cold. He lasted almost an hour without shivering; when my turn comes, I shiver aggressively on the cot for nearly an hour straight. I'm visiting Texas to learn about this experiment on how different bodies respond to extreme climates. I jokingly ask Cowgill as she tapes biosensing devices to my chest and legs. After I exit the cold, she surprises me: "You, believe it or not, were not the worst person we've ever seen." Climate change forces us to reckon with the knotty science of how our bodies interact with the environment. Cowgill is a 40-something anthropologist at the University of Missouri who powerlifts and teaches CrossFit in her spare time. She's small and strong, with dark bangs and geometric tattoos. Since 2022, she's spent the summers at the University of North Texas Health Science Center tending to these uncomfortable experiments. Her team hopes to revamp the science of thermoregulation. While we know in broad strokes how people thermoregulate, the science of keeping warm or cool is mottled with blind spots. "We have the general picture.

The 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025) will be held from 19-25 October in Hangzhou, China. The programme includes plenary and keynote talks, workshops, tutorials, forums, competitions, and a debate. There are three plenary talks on the programme this year, with one per day on Tuesday 21, Wednesday 22, and Thursday 23 October. On Wednesday, a debate will be held on the following topic: The participants will be: XingXing Wang (Unitree Robotics), Jun-Oh Ho (Samsung and Rainbow Robotics), Hong Qiao (Chinese Academy of Sciences), Andra Keay, (Silicon Valley Robotics), Yu Sun (EiC, IEEE Trans on Automation Science and Engineering), Tamim Asfour (Professor of Humanoid Robotics, Karlsruhe Institute of Technology), Ken Goldberg (UC Berkeley, Moderator). There are three tutorials planned, taking place on Monday 20 and Friday 24 October.

Stochastic Gradient Descent (SGD) and its Ruppert-Polyak averaged variant (ASGD) lie at the heart of modern large-scale learning, yet their theoretical properties in high-dimensional settings are rarely understood. In this paper, we provide rigorous statistical guarantees for constant learning-rate SGD and ASGD in high-dimensional regimes. Our key innovation is to transfer powerful tools from high-dimensional time series to online learning. Specifically, by viewing SGD as a nonlinear autoregressive process and adapting existing coupling techniques, we prove the geometric-moment contraction of high-dimensional SGD for constant learning rates, thereby establishing asymptotic stationarity of the iterates. Building on this, we derive the $q$-th moment convergence of SGD and ASGD for any $q\ge2$ in general $\ell^s$-norms, and, in particular, the $\ell^{\infty}$-norm that is frequently adopted in high-dimensional sparse or structured models. Furthermore, we provide sharp high-probability concentration analysis which entails the probabilistic bound of high-dimensional ASGD. Beyond closing a critical gap in SGD theory, our proposed framework offers a novel toolkit for analyzing a broad class of high-dimensional learning algorithms.

Effective feedback is essential for student learning but is time-intensive for teachers. We present LearnLens, a modular, LLM-based system that generates personalised, curriculum-aligned feedback in science education. LearnLens comprises three components: (1) an error-aware assessment module that captures nuanced reasoning errors; (2) a curriculum-grounded generation module that uses a structured, topic-linked memory chain rather than traditional similarity-based retrieval, improving relevance and reducing noise; and (3) an educator-in-the-loop interface for customisation and oversight. LearnLens addresses key challenges in existing systems, offering scalable, high-quality feedback that empowers both teachers and students.

Speculative decoding generally dictates having a small, efficient draft model that is either pretrained or distilled offline to a particular target model series, for instance, Llama or Qwen models. However, within online deployment settings, there are two major challenges: 1) usage of a target model that is incompatible with the draft model; 2) expectation of latency improvements over usage and time. In this work, we propose OmniDraft, a unified framework that enables a single draft model to operate with any target model and adapt dynamically to user data. We introduce an online n-gram cache with hybrid distillation fine-tuning to address the cross-vocabulary mismatch across draft and target models; and further improve decoding speed by leveraging adaptive drafting techniques. OmniDraft is particularly suitable for on-device LLM applications where model cost, efficiency and user customization are the major points of contention. This further highlights the need to tackle the above challenges and motivates the \textit{``one drafter for all''} paradigm. We showcase the proficiency of the OmniDraft framework by performing online learning on math reasoning, coding and text generation tasks. Notably, OmniDraft enables a single Llama-68M model to pair with various target models including Vicuna-7B, Qwen2-7B and Llama3-8B models for speculative decoding; and additionally provides up to 1.5-2x speedup.

Dense retrievers play a vital role in accessing external and specialized knowledge to augment language models (LMs). Training dense retrievers typically requires annotated query-document pairs, which are costly to create and scarce in specialized domains (e.g., code) or in complex settings (e.g., requiring reasoning). These practical challenges have sparked growing interest in self-supervised retriever learning. Since LMs are trained to capture token-level dependencies through a self-supervised learning objective (i.e., next token prediction), we can analogously cast retrieval as learning dependencies among chunks of tokens. This analogy naturally leads to the question: How can we adapt self-supervised learning objectives in the spirit of language modeling to train retrievers? To answer this question, we introduce Revela, a unified and scalable training framework for self-supervised retriever learning via language modeling. This attention is weighted by retriever-computed similarity scores, enabling the retriever to be optimized as part of language modeling. We evaluate Revela on domain-specific (CoIR), reasoning-intensive (BRIGHT), and general-domain (BEIR) benchmarks across various retriever backbones. Without annotated or synthetic query-document pairs, Revela surpasses larger supervised models and proprietary APIs on CoIR and matches them on BRIGHT. It achieves BEIR's unsupervised SoT A with 1000x less training data and 10x less compute. Central to information retrieval are dense retrievers (Reimers & Gurevych, 2019; Karpukhin et al., 2020; Ma et al., 2024), which map queries and documents into high-dimensional vector spaces and determine relevance through similarity calculations. Typically, these models rely on carefully annotated query-document pairs and hard negatives for training.

As LLMs become central to interactive applications, ranging from tutoring to mental health, the ability to express personality in culturally appropriate ways is increasingly important. While recent works have explored personality evaluation of LLMs, they largely overlook the interplay between culture and personality. To address this, we introduce CulturalPersonas, the first large-scale benchmark with human validation for evaluating LLMs' personality expression in culturally grounded, behaviorally rich contexts. Our dataset spans 3,000 scenario-based questions across six diverse countries, designed to elicit personality through everyday scenarios rooted in local values. We evaluate three LLMs, using both multiple-choice and open-ended response formats. Our results show that CulturalPersonas improves alignment with country-specific human personality distributions (over a 20% reduction in Wasserstein distance across models and countries) and elicits more expressive, culturally coherent outputs compared to existing benchmarks. CulturalPersonas surfaces meaningful modulated trait outputs in response to culturally grounded prompts, offering new directions for aligning LLMs to global norms of behavior. By bridging personality expression and cultural nuance, we envision that CulturalPersonas will pave the way for more socially intelligent and globally adaptive LLMs.