From 10th to 12th September 2025, Barcelona hosted an academic gathering at the Universitat Oberta de Catalunya: the first Hype Studies Conference, titled "(Don't) Believe the Hype!?" Organised by a transnational, collective research group of scholars and practitioners, the conference drew together researchers, activists, artists, journalists, and technology professionals to examine hype as a significant force shaping contemporary society. Hype Studies is an emerging academic field that analyses how and why excessive expectations form around technologies, ideas, or phenomena, and what effects those expectations have on society, culture, economics, and policy. As the playful brackets around "Don't" in the conference title suggest - both a warning and an invitation to question that warning - the aim of the conference wasn't to simply reject hype, but to understand it. The conference approached hype critically by examining it as a phenomenon with real power and consequences that needs to be understood and questioned. The purpose here was to build collective knowledge about hype, develop better and more concrete theories, share empirical findings, and create an interdisciplinary community whilst advancing the field's scholarship and knowledge.

David Ha, CEO of Sakana AI, is one of the founders of the startup, which was established in 2023. Sakana AI has achieved a ¥400 billion ($2.6 billion) valuation after raising ¥20 billion yen in a recent funding round, and notes that it is now Japan's largest startup ever. The new investment and the sky-high valuation come as AI-mania grips markets globally and drives shares of publicly traded AI-related companies to unprecedented levels. Nvidia now has a market capitalization of more than $4.5 trillion. It also comes as more questions are raised about AI and whether it will deliver on its promises and generate enough profit to justify all the capital being committed to it. In a time of both misinformation and too much information, quality journalism is more crucial than ever.

There are no catchy melodies in McCartney's first release in five years, only quiet hiss and the odd clatter. There are no catchy melodies in McCartney's first release in five years, only quiet hiss and the odd clatter. But Paul McCartney's first new recording in five years lacks the sing-along tune and jaunty guitar chops because there's barely anything there. In place of catchy melodies and evocative lyrics there is only quiet hiss and the odd clatter. It suggests that if AI companies unfairly exploit musicians' intellectual property to train their generative AI models, the creative ecosystem will be wrecked and original music silenced.

B ased on a true story, Bill Guttentag's rousing drama attests to the resilience of women who dare to dream despite draconian social strictures. The film follows Roya Mahboob (Nikohl Boosheri), a trailblazing coach and businesswoman in Stem (science, technology, engineering and mathematics) who assembles a robotics team of Afghan girls for international competitions. They face the same dangers too; in a country where women are not encouraged or even allowed to pursue higher levels of education, their quest for medals sees opposition from their own families as well as public scorn from conservatives. Rule Breakers is at its most thrilling during the competition sequences, which splice together real-life documentary footage of the events with fictional re-enactments. These spaces are portrayed as a haven that encourages camaraderie rather than competitiveness, and in a world divided by military conflicts and war, they offer a utopiian vision of international collaboration and solidarity.

Mysterious drones have been spotted at night at airports across Europe. How worried should we be? First comes the warning, that disembodied voice over the tannoy: Your attention please. Please move to the shelter on the minus second floor. Then comes the mosquito-like whine of the incoming Russian drones, massing in their hundreds just above the clouds.

How my on-air'brain fog' moment sparked a big debate When I rather nervously shared a personal post about dealing with brain fog at work on the social network LinkedIn last week, I had no idea that it would have such an enormous impact. It's been viewed hundreds of thousands of times. Women have stopped me on the street to talk to me about it. I've been overwhelmed by hundreds of messages from people sharing support and their own experiences of it. Usually I cover technology news.

Direct Preference Optimization (DPO) has emerged as a simple alternative to reinforcement learning from human feedback (RLHF) for aligning language models, but its reliance on hard pairwise labels makes it brittle under noise; our experiments show performance degrading by up to 93 percent in noisy settings. We introduce Anchored Direct Preference Optimization (ADPO), a unified framework that addresses this fragility through reference anchoring. By minimizing KL(q || softmax((l - l_ref) / tau_anc)), where l_ref are reference policy log probabilities, ADPO provides three key advantages: (1) it unifies major learning paradigms, including supervised fine-tuning, knowledge distillation, maximum-entropy reinforcement learning, and DPO, as special cases through different choices of target distribution q, anchor policy pi_ref, and temperature tau_anc; (2) it induces an implicit trust region governed by the softmax Fisher metric with curvature scaling as 1 / tau_anc^2, providing geometric regularization absent in standard methods; and (3) it enables flexible anchor strategies tailored to different learning contexts. Empirically, ADPO consistently outperforms standard DPO by 12 to 93 percent across twelve noisy scenarios, with listwise variants achieving top performance in eleven of twelve cases. In offline distillation, ADPO reduces student-teacher KL by 4 to 49 times while achieving superior returns (for example, 279.3 vs -309.0 for knowledge distillation on HalfCheetah). We further uncover a task-dependent tradeoff: dynamic anchors excel at online exploration in noisy environments (plus 5 to 11 percent), while fixed anchors enable stable offline distillation. Our work establishes anchoring as a general principle for robust policy optimization, with clear practical guidance for anchor selection across diverse learning scenarios.

Linear models are widely used in high-stakes decision-making due to their simplicity and interpretability. Yet when fairness constraints such as demographic parity are introduced, their effects on model coefficients, and thus on how predictive bias is distributed across features, remain opaque. Existing approaches on linear models often rely on strong and unrealistic assumptions, or overlook the explicit role of the sensitive attribute, limiting their practical utility for fairness assessment. We extend the work of (Chzhen and Schreuder, 2022) and (Fukuchi and Sakuma, 2023) by proposing a post-processing framework that can be applied on top of any linear model to decompose the resulting bias into direct (sensitive-attribute) and indirect (correlated-features) components. Our method analytically characterizes how demographic parity reshapes each model coefficient, including those of both sensitive and non-sensitive features. This enables a transparent, feature-level interpretation of fairness interventions and reveals how bias may persist or shift through correlated variables. Our framework requires no retraining and provides actionable insights for model auditing and mitigation. Experiments on both synthetic and real-world datasets demonstrate that our method captures fairness dynamics missed by prior work, offering a practical and interpretable tool for responsible deployment of linear models.

We study randomized experiments in bipartite systems where only a subset of treatment-side units are eligible for assignment while all units continue to interact, generating interference. We formalize eligibility-constrained bipartite experiments and define estimands aligned with full deployment: the Primary Total Treatment Effect (PTTE) on eligible units and the Secondary Total Treatment Effect (STTE) on ineligible units. Under randomization within the eligible set, we give identification conditions and develop interference-aware ensemble estimators that combine exposure mappings, generalized propensity scores, and flexible machine learning. We further introduce a projection that links treatment- and outcome-level estimands; this mapping is exact under a Linear Additive Edges condition and enables estimation on the (typically much smaller) treatment side with deterministic aggregation to outcomes. In simulations with known ground truth across realistic exposure regimes, the proposed estimators recover PTTE and STTE with low bias and variance and reduce the bias that could arise when interference is ignored. Two field experiments illustrate practical relevance: our method corrects the direction of expected interference bias for a pre-specified metric in both studies and reverses the sign and significance of the primary decision metric in one case.

One of the major bottlenecks for deploying popular first-order differentially private (DP) machine learning algorithms (e.g., DP-SGD) lies in their high computation and memory cost, despite the existence of optimized implementations. Zeroth-order methods have promise in mitigating the overhead, as they leverage function evaluations to approximate the gradients, hence significantly easier to privatize. While recent works have explored zeroth-order approaches in both private and non-private settings, they still suffer from relatively low utilities compared with DP-SGD, and have only been evaluated in limited application domains. In this work, we propose to leverage public information to guide and improve gradient approximation of private zeroth-order algorithms. We explore a suite of public-data-assisted zeroth-order optimizers (PAZO) with minimal overhead. We provide theoretical analyses of the PAZO framework under an assumption of the similarity between public and private data. Empirically, we demonstrate that PAZO achieves superior privacy/utility tradeoffs across vision and text tasks in both pre-training and fine-tuning settings, outperforming the best first-order baselines (with public data) especially in highly private regimes, while offering up to $16\times$ runtime speedup.