Understanding how deep neural networks learn useful internal representations from data remains a central open problem in the theory of deep learning. We introduce Neural Low-Degree Filtering (Neural LoFi), a stylized limit of gradient-based training in which hierarchical feature learning becomes an explicit iterative spectral procedure. In this limit, the dynamics at each layer decouple: given the current representation, the next layer selects directions with maximal accessible low-degree correlation to the label. This yields a tractable surrogate mechanism for deep learning, together with a natural kernel-space interpretation. Neural LoFi provides a mathematically explicit framework for studying multi-layer feature learning beyond the lazy regime. It predicts how representations are selected layer by layer, explains how emergence of concepts arises with given sample complexity,and gives a concrete mechanism by which depth progressively constructs new features from old ones through low-degree compositionality. We complement the theory with mechanistic experiments on fully connected and convolutional architectures, showing that Neural LoFi improves over lazy random-feature baselines, recovers meaningful structured filters, and predicts representations aligned with early gradient-descent feature discovery with real datasets.

We introduce a family of synthetic languages with hierarchical structure -- generated by a broadcast process on trees -- for which the role of context length and reasoning in autoregressive generation can be analyzed precisely. At the heart of our analytic approach is an \emph{exact $k$-gram ansatz} in place of transformers with context length $k$, a substitution we then validate empirically. Using this ansatz we derive explicit asymptotic predictions for distributional statistics of the sequences produced by a trained model, instantiated in two settings. For the \emph{Ising broadcast process} (a soft-constrained language), we prove that the variance of the generated sum scales log-linearly in the context depth and its kurtosis converges to that of a Gaussian -- both deviating from the true language for any sublinear context. For the \emph{coloring broadcast process} (a hard-constrained language) in the freezing regime, bounded-context autoregression produces sequences that, with high probability, are inconsistent with \emph{any} valid coloring of the underlying tree. Together these results imply an $Ω(n)$ lower bound on the context length required to faithfully sample length-$n$ sequences. In contrast, we prove that an autoregressive \emph{reasoning} model with only $Θ(\log n)$ working memory can sample exactly from the true language -- an exponential improvement. We confirm both the lower-bound predictions and the reasoning-based upper bound empirically with transformers trained on the synthetic language; the trained models track our asymptotic predictions quantitatively across a wide range of context sizes.

OpenAI, which is currently facing a raft of lawsuits over alleged safety lapses in ChatGPT, has endorsed the Kids Online Safety Act (KOSA). The company said that its endorsement was part of a broader commitment to create AI-specific rules for kids safety. OpenAI's endorsement comes as KOSA, which passed the Senate in 2024, appears to be gaining some momentum . KOSA, which was first introduced in 2022, is one of several online safety bills that would require social media companies and other online platforms to implement stronger protections for children. The bill has been revised a number of times, but the current version includes a requirement for social media apps to allow minors to opt out of addictive features and algorithmic recommendations.

AI chatbots are giving out people's real phone numbers People report that their personal contact info was surfaced by Google AI--and there's apparently no easy way to prevent it. A Redditor recently wrote that he was "desperate for help": for about a month, he said, his phone had been inundated by calls from "strangers" who were "looking for a lawyer, a product designer, a locksmith." Callers were apparently misdirected by Google's generative AI. In March, a software developer in Israel was contacted on WhatsApp after Google's chatbot Gemini provided incorrect customer service instructions that included his number. And in April, a PhD candidate at the University of Washington was messing around on Gemini and got it to cough up her colleague's personal cell phone number. AI researchers and online privacy experts have long warned of the myriad dangers generative AI poses for personal privacy.

In, the company sought to show the jury a remarkable trophy as physical proof of Elon Musk's concerning behavior. Wednesday's episode of the trial kicked off on Wednesday with a unique proposition: OpenAI wanted to bring its ass into the courtroom, and lay it bare before the jury. It's a good thing lady justice wears that blindfold. A lawyer for Sam Altman's AI behemoth, Bradley Wilson, approached US district judge Yvonne Gonzalez Rogers and handed her a small gold statue with a white stone base. It depicted the rear end of a donkey--with two legs, a butt, and a tail--and was inscribed with the message, "Never stop being a jackass for safety."

The 10th International Workshop on Health Intelligence (W3PHIAI-26) celebrated a decade of bringing AI and health research together, building on a lineage that began with the AAAI-W3PHI workshops focused on population health (2014-2016), the AAAI-HIAI workshops focused on personalized health (2013-2016), and the subsequent joint W3PHIAI workshops held annually from 2017 through 2025. Over this decade, the series has produced hundreds of talks and high-impact publications that have collectively received thousands of citations, shaping the research agenda in both population health intelligence and personalized healthcare AI. This year's special theme, "Foundation Models and AI Agents," reflected the field's rapidly evolving frontier: the emergence of autonomous and semi-autonomous AI systems reshaping clinical workflows, patient management, health system operations, and public health surveillance. Day 1 of the workshop focused on medical imaging and the translation of AI for clinical ...

The company says its new Incognito Chat allows you to use its AI chatbot without anyone else--including Meta--being able to access your conversations. WhatsApp said on Wednesday it is launching an AI chat function known as Incognito Chat that is built to allow users to converse privately with Meta AI --such that Meta itself cannot access the questions or answers. The feature is based on WhatsApp's Private Processing scheme, which debuted a year ago and already underlies WhatsApp's existing AI features, including message summarization and composition tools. The idea of Incognito Chat is to create a way for WhatsApp to offer AI chat integration that does not conflict with the communication platform's commitment to end-to-end encryption, the privacy scheme in which only direct participants in a conversation can read messages or hear a call. Most generative AI platforms now offer some type of "incognito mode," but these features are usually designed to separate users from the questions they ask and the answers they receive rather than including a mechanism to entirely shield those questions and answers from the provider's view.

Plus: Sam Altman claims Elon Musk tried to seize control of OpenAI. A startup called Varda Space Industries is betting that the future of pharmaceuticals lies in orbit. The company has signed a deal with United Therapeutics to test whether drugs crystallize differently in microgravity, potentially creating improved versions with new properties. The idea sounds futuristic, but falling launch costs and reusable rockets are making space-based manufacturing seem increasingly plausible. Varda says the partnership could mark an important step toward building products in orbit for use back on Earth. Discover how space could become the next frontier for drug development .

Government departments and other public bodies in the UK must consider requests to release information about AI-produced content, regulators have confirmed. The move follows a successful request by New Scientist for the release of a minister's ChatGPT logs The use of AI chatbots is subject to the UK's Freedom of Information laws Text, images and other content produced by UK government departments and other public bodies using artificial intelligence are subject to freedom of information (FOI) laws, regulators have confirmed - potentially opening the door for the public to gain access to ministers' ChatGPT or other chatbot records. The Information Commissioner's Office (ICO), the UK's data-protection agency, has released new guidance confirming that "If staff at a public authority use AI for work purposes, the information generated will be subject to FOIA [the Freedom of Information Act] along with the prompts used". Last year, successfully requested the then-UK tech secretary Peter Kyle's ChatGPT logs under FOI legislation, in what is believed to be a world first. That triggered subsequent requests from other news outlets to obtain other information, but many have either been rejected on cost grounds or labelled as "vexatious", an umbrella term that allows authorities to reject a request.

Beware what you tell your AI chatbot. It's not a shrink - it's a snitch In a case of'oh dear diary', the OpenAI president Greg Brockman is having to read extracts from his musings about Elon Musk in court. T he hottest new read of 2026 may well be The Secret Diary of Greg Brockman, Aged 38 . It's got everything: feuding billionaires, scheming CEOs and a perhaps somewhat unreliable narrator. You won't find it in the library, but you can watch Brockman, a co-founder and president of OpenAI, being forced to read the juiciest bits out loud in court. Before you ask ChatGPT to explain, here's the backstory: Elon Musk is in a legal battle with Brockman and the OpenAI CEO, Sam Altman .