Don't make us security guards, says teacher stabbed by pupil A teacher who thought she was going to die when she was stabbed by a 13-year-old pupil in the schoolyard has said giving staff handheld scanners will not stop violence in schools. Liz Hopkin, who was attacked at Ysgol Dyffryn Aman in 2024, said she felt really worried after the Welsh government announced it would offer school staff more guidance on what to do if they suspected a pupil had brought a weapon into school. It comes as a 15-year-old boy was charged with attempted murder after a teacher was stabbed at a school in the neighbouring county. Hopkin said teachers aren't security, while the Welsh government said the resources were about prevention, building on existing guidance. Hopkin, her colleague Fiona Elias and a pupil were attacked at the school where she worked in Ammanford, Carmarthenshire, by a girl who had previously been found with a knife.

What does it mean to have speech without a speaker? For the first time, speech has been decoupled from consequence. We now live alongside AI systems that converse knowledgeably and persuasively--deploying claims about the world, explanations, advice, encouragement, apologies, and promises--while bearing no vulnerability for what they say. Millions of people already rely on chatbots powered by large language models, and have integrated these synthetic interlocutors into their personal and professional lives. An LLM's words shape our beliefs, decisions, and actions, yet no speaker stands behind them. This dynamic is already familiar in everyday use. A chatbot gets something wrong. When corrected, it apologizes and changes its answer.

Transformers empirically perform precise probabilistic reasoning in carefully constructed ``Bayesian wind tunnels'' and in large-scale language models, yet the mechanisms by which gradient-based learning creates the required internal geometry remain opaque. We provide a complete first-order analysis of how cross-entropy training reshapes attention scores and value vectors in a transformer attention head. Our core result is an \emph{advantage-based routing law} for attention scores, \[ \frac{\partial L}{\partial s_{ij}} = α_{ij}\bigl(b_{ij}-\mathbb{E}_{α_i}[b]\bigr), \qquad b_{ij} := u_i^\top v_j, \] coupled with a \emph{responsibility-weighted update} for values, \[ Δv_j = -η\sum_i α_{ij} u_i, \] where $u_i$ is the upstream gradient at position $i$ and $α_{ij}$ are attention weights. These equations induce a positive feedback loop in which routing and content specialize together: queries route more strongly to values that are above-average for their error signal, and those values are pulled toward the queries that use them. We show that this coupled specialization behaves like a two-timescale EM procedure: attention weights implement an E-step (soft responsibilities), while values implement an M-step (responsibility-weighted prototype updates), with queries and keys adjusting the hypothesis frame. Through controlled simulations, including a sticky Markov-chain task where we compare a closed-form EM-style update to standard SGD, we demonstrate that the same gradient dynamics that minimize cross-entropy also sculpt the low-dimensional manifolds identified in our companion work as implementing Bayesian inference. This yields a unified picture in which optimization (gradient flow) gives rise to geometry (Bayesian manifolds), which in turn supports function (in-context probabilistic reasoning).

'Ridiculous amount of games' - has Haaland played too much football? The robot is malfunctioning and in need of a reset. Erling Haaland made a blistering start to the season but that prolific run of form has suffered a glitch. Though the Manchester City and Norway striker has scored a remarkable 39 goals in just 36 games for club and country this season, he has hit a sticky patch of form with only one goal in his past eight games. This has coincided with Pep Guardiola's men falling off the pace in the Premier League title race and suffering a monumental shock at Bodo/Glimt in the Champions League.

Gradient descent on log-sum-exp (LSE) objectives performs implicit expectation--maximization (EM): the gradient with respect to each component output equals its responsibility. The same theory predicts collapse without volume control analogous to the log-determinant in Gaussian mixture models. We instantiate the theory in a single-layer encoder with an LSE objective and InfoMax regularization for volume control. Experiments confirm the theory's predictions. The gradient--responsibility identity holds exactly; LSE alone collapses; variance prevents dead components; decorrelation prevents redundancy. The model exhibits EM-like optimization dynamics in which lower loss does not correspond to better features and adaptive optimizers offer no advantage. The resulting decoder-free model learns interpretable mixture components, confirming that implicit EM theory can prescribe architectures.

Machine Learning (ML) has been a foundational topic in artificial intelligence (AI), providing both theoretical groundwork and practical tools for its exciting advancements. From ResNet for visual recognition to Transformer for vision-language alignment, the AI models have achieved superior capability to humans. Furthermore, the scaling law has enabled AI to initially develop general intelligence, as demonstrated by Large Language Models (LLMs). To this stage, AI has had an enormous influence on society and yet still keeps shaping the future for humanity. However, distribution shift remains a persistent ``Achilles' heel'', fundamentally limiting the reliability and general usefulness of ML systems. Moreover, generalization under distribution shift would also cause trust issues for AIs. Motivated by these challenges, my research focuses on \textit{Trustworthy Machine Learning under Distribution Shifts}, with the goal of expanding AI's robustness, versatility, as well as its responsibility and reliability. We carefully study the three common distribution shifts into: (1) Perturbation Shift, (2) Domain Shift, and (3) Modality Shift. For all scenarios, we also rigorously investigate trustworthiness via three aspects: (1) Robustness, (2) Explainability, and (3) Adaptability. Based on these dimensions, we propose effective solutions and fundamental insights, meanwhile aiming to enhance the critical ML problems, such as efficiency, adaptability, and safety.

The head of Amazon Web Services has big plans to offer AI tools to businesses, but says that replacing coders with AI is "a non-starter for anyone who's trying to build a long-term company." Amid the breathless coverage and relentless AI hype of recent years, one of the world's biggest tech companies--Amazon--has been notably absent. Matt Garman, the CEO of Amazon Web Services, is looking to change that. At the recent AWS re:Invent conference, Garman announced a bunch of frontier AI models, as well as a tool designed to let AWS customers build models of their own. That tool, Nova Forge, allows companies to engage in what's known as custom pretraining--adding their data in the process of building a base model--which should allow for vastly more customized models that suit a given company's needs. Sure, it doesn't quite have the sexiness of a Sora 2 announcement, but that's not Garman's goal: He's less interested in mass consumer use of AI and more interested in enterprise solutions that'll integrate AI into all of AWS's offerings--and have a material impact on a corporate P&L. For this week's episode of, I caught up with Garman after AWS re:Invent to talk about what the company announced, whether he feels behind in the AI race, how he thinks about managing huge teams (and managing internal dissent), and why he's not convinced that AI is (or should be) the great job thief of our era. We always start these conversations with some very quick questions, like a warmup. If AWS had a mascot, what would it be? We have a big S3 bucket sometimes that goes around, so we'll call it that. Sorry, what is an S3 bucket? An S3 bucket is like a thing that you store your S3 objects in, but we actually have a large foam big bucket that walks around and actually looks like a paint bucket. So you do have a mascot. Well, S3 has a bucket, it has a mascot. It's probably the closest we have, and I like it. What's the most expensive mistake you've ever made? Personally, the most expensive mistake I ever made was playing basketball too long and I tore my Achilles. So that cost me about nine months of being able to walk. I probably should have known that into my thirties I was well past basketball-playing age.

To counter fragmented, high-risk adoption of commercial AI tools, we built and ran an institutional AI platform in a six-month, 300-user pilot, showing that a university of applied sciences can offer advanced AI with fair access, transparent risks, controlled costs, and alignment with European law. Commercial AI subscriptions create unequal access and compliance risks through opaque processing and non-EU hosting, yet banning them is neither realistic nor useful. Institutions need a way to provide powerful AI in a sovereign, accountable form. Our solution is a governed gateway platform with three layers: a ChatGPT-style frontend linked to institutional identity that makes model choice explicit; a gateway core enforcing policy, controlling access and budgets, and routing traffic to EU infrastructure by default; and a provider layer wrapping commercial and open-source models in institutional model cards that consolidate vendor documentation into one governance interface. The pilot ran reliably with no privacy incidents and strong adoption, enabling EU-default routing, managed spending, and transparent model choices. Only the gateway pattern combines model diversity and rapid innovation with institutional control. The central insight: AI is not a support function but strategy, demanding dedicated leadership. Sustainable operation requires governance beyond traditional boundaries. We recommend establishing a formal AI Officer role combining technical literacy, governance authority, and educational responsibility. Without it, AI decisions stay ad-hoc and institutional exposure grows. With it, higher-education institutions can realistically operate their own multi-provider AI platform, provided they govern AI as seriously as they teach it.

The Endless Tuning is a design method for a reliable deployment of artificial intelligence based on a double mirroring process, which pursues both the goals of avoiding human replacement and filling the so-called responsibility gap (Matthias 2004). Originally depicted in (Fabris et al. 2024) and ensuing the relational approach urged therein, it was then actualized in a protocol, implemented in three prototypical applications regarding decision-making processes (respectively: loan granting, pneumonia diagnosis, and art style recognition) and tested with such as many domain experts. Step by step illustrating the protocol, giving insights concretely showing a different voice (Gilligan 1993) in the ethics of artificial intelligence, a philosophical account of technical choices (e.g., a reversed and hermeneutic deployment of XAI algorithms) will be provided in the present study together with the results of the experiments, focusing on user experience rather than statistical accuracy. Even thoroughly employing deep learning models, full control was perceived by the interviewees in the decision-making setting, while it appeared that a bridge can be built between accountability and liability in case of damage.