Since 2021, China has reportedly poured $100bn into support for AI datacentres. Since 2021, China has reportedly poured $100bn into support for AI datacentres. Beijing's AI policy is focused on real-life applications but Chinese companies are beginning to articulate their own grand visions S tanding on stage in the eastern China tech hub of Hangzhou, Alibaba's normally media-shy CEO made an attention-grabbing announcement. "The world today is witnessing the dawn of an AI-driven intelligent revolution," Eddie Wu told a developer conference in September. " Artificial general intelligence (AGI) will not only amplify human intelligence but also unlock human potential, paving the way for the arrival of artificial superintelligence (ASI)."

What AI "remembers" about you is privacy's next frontier Agents' technical underpinnings create the potential for breaches that expose the entire mosaic of your life. The ability to remember you and your preferences is rapidly becoming a big selling point for AI chatbots and agents. Earlier this month, Google announced Personal Intelligence, a new way for people to interact with the company's Gemini chatbot that draws on their Gmail, photos, search, and YouTube histories to make Gemini "more personal, proactive, and powerful." It echoes similar moves by OpenAI, Anthropic, and Meta to add new ways for their AI products to remember and draw from people's personal details and preferences. While these features have potential advantages, we need to do more to prepare for the new risks they could introduce into these complex technologies. Personalized, interactive AI systems are built to act on our behalf, maintain context across conversations, and improve our ability to carry out all sorts of tasks, from booking travel to filing taxes.

Plus: TikTok won't be heading to court this week The first human test of a rejuvenation method will begin "shortly" Life Biosciences, a small Boston startup founded by Harvard professor and life-extension evangelist David Sinclair, has won FDA approval to proceed with the first targeted attempt at age reversal in human volunteers. The company plans to try to treat eye disease with a radical rejuvenation concept called "reprogramming" that has recently attracted hundreds of millions in investment for Silicon Valley firms like Altos Labs, New Limit, and Retro Biosciences, backed by many of the biggest names in tech. Today, an estimated 2.2 billion people still have either limited or no access to the internet, largely because they live in remote places. But that number could drop this year, thanks to tests of stratospheric airships, uncrewed aircraft, and other high-altitude platforms for internet delivery. Although Google shuttered its high-profile internet balloon project Loon in 2021, work on other kinds of high-altitude platform stations has continued behind the scenes. Now, several companies claim they have solved Loon's problems--and are getting ready to prove the tech's internet beaming potential starting this year.

'This train isn't going to stop': shocking Sundance film shows promises and perils of AI Is AI an existential threat, or an epochal opportunity? Those are the questions top of mind for a new documentary at Sundance, which features leading AI experts, critics and entrepreneurs, including Sam Altman, the OpenAI CEO, with views on the near-to-midterm future ranging from doom to utopia. 'The world is hurting right now': politics and protest hit the Sundance film festival The AI Doc: Or How I Became an Apocaloptimist, directed by Daniel Roher and Charlie Tyrell and produced by Daniel Kwan (one half of The Daniels, the Oscar-winning duo behind Everything Everywhere All At Once), delves into the contentious topic of AI through Roher's own anxiety. The Canadian film-maker, who won an Oscar in 2023 for the documentary Navalny, first became interested in the topic while experimenting with tools released by OpenAI, the company behind the chatbot ChatGPT. The sophistication of the public tools - the ability to produce whole paragraphs in seconds, or produce illustrations - both thrilled and unnerved him.

SoftBank Group is in discussions to invest as much as $30 billion more in OpenAI, a sharp increase in commitment that reflects founder Masayoshi Son's ambitions to play a central role in developing artificial intelligence. The Japanese company, already one of the ChatGPT-maker's biggest backers, is in deliberations to commit more capital toward the fast-growing startup, people familiar with the matter said. The maximum amount SoftBank is considering is $30 billion, one of the people said, asking to remain anonymous to discuss private talks. They cautioned that the discussions are fluid and the amount of funding could change. SoftBank's shares rose 5.8% in Tokyo on Wednesday. Son has been unwinding positions to increase its stake in OpenAI and ready capital for sweeping investments aimed at injecting AI in all devices.

Devastating impact of Minneapolis shooting on Trump is worse than expected: Poll reveals America's crushing verdict... and what he must do next Bodies are STILL in wreckage of private jet that crashed in Maine on Sunday, killing six including powerful lawyer's attorney wife School principal accused of shoplifting from Walmart using'stacking' method at self-checkout Melania's shock role in Trump's showdown with Kristi Noem revealed: MARK HALPERIN's fly-on-wall account of Oval Office meeting... and who is ACTUALLY taking the fall for Alex Pretti shooting I was barely eating but kept gaining weight. Then I discovered the'taboo' cancer doctors NEVER talk about. Now sex will never be the same... don't ignore these signs Harper Beckham, 14, puts on a stylish display in a fluffy coat and vintage Chanel bag in Paris with her family - after Nicola Peltz's heartbreaking comments about sister-in-law Devastating truth about Blind Side actor Quinton Aaron: More to this'than everyone is letting on', friends reveal... as co-star Sandra Bullock'monitors' situation The wild truth about my influencer sons, their psycho dad and how lawsuits nearly left them bankrupt - by Jake and Logan Paul's MOM Trump knifes'little Napoleon' Border Patrol commander over Minnesota mayhem as he declares: 'We'll de-escalate' Lost tomb of the mysterious'cloud people' unearthed after 1,400 years in'discovery of the decade' READ MORE: How AI cops will be used to patrol Britain's streets Scientists have uncovered four entirely new personality types that all ChatGPT users fall into. According to experts from the University of Oxford and the Berlin University Alliance, every chatbot user has a unique personality type, each with their own motivations. Some truly tech-savvy users fall into the category of ' AI enthusiasts'.

We propose a framework to construct practical kernel-based two-sample tests from the family of $f$-divergences. The test statistic is computed from the witness function of a regularized variational representation of the divergence, which we estimate using kernel methods. The proposed test is adaptive over hyperparameters such as the kernel bandwidth and the regularization parameter. We provide theoretical guarantees for statistical test power across our family of $f$-divergence estimates. While our test covers a variety of $f$-divergences, we bring particular focus to the Hockey-Stick divergence, motivated by its applications to differential privacy auditing and machine unlearning evaluation. For two-sample testing, experiments demonstrate that different $f$-divergences are sensitive to different localized differences, illustrating the importance of leveraging diverse statistics. For machine unlearning, we propose a relative test that distinguishes true unlearning failures from safe distributional variations.

We study how to fine-tune LLMs using user-edit deployment data consisting of a set of context, an agent's response, and user edits. This deployment data is naturally generated by users in applications such as LLMs-based writing assistants and coding agents. The _natural_ origin of user edits makes it a desired source for adapting and personalizing LLMs. In this setup, there emerges a unification of various feedback types namely preferences, supervised labels, and cost that are typically studied separately in the literature. In this paper, we initiate the theoretical investigation of learning from user edits. We first derive bounds for learning algorithms that learn from each of these feedback types. We prove that these algorithms have different trade-offs depending upon the user, data distribution, and model class. We then propose a simple ensembling procedure to jointly learn from these feedback types. On two domains adapted from Gao et al. 2024, we show our ensembling procedure outperforms these methods that learn from individual feedback. Further, we show that our proposed procedure can robustly adapt to different user-edit distributions at test time.

Recommender systems have become ubiquitous across a wide range of fields, such as ecommerce, media consumption (including movies, books, music, news, etc.), social networks, finance, and many others, due to their effectiveness in identifying relevant items or content among numerous choices [1, 2]. Traditionally, recommender systems, largely based on collaborative filtering techniques, have focused on recommending individual (or "atomic") items, such as movies or books, by understanding users' preferences for these individual items. However, in certain application domains, recommending "composite" items (i.e., combinations of atomic items) represents a very important capability. For illustration, consider a clothing/fashion recommender system, where we want to recommend "outfits" - combinations of tops (t-shirts, shirts, sweaters) and bottoms (pants, skirts, shorts) - to users. In such a case, multiple fashion items in a recommended outfit ideally have to match both functionally and stylistically, which may require domain expertise (e.g., on things like style compatibility) beyond individual preferences. Another key challenge for such recommender systems is that a given user's personal preference for a composite item may not directly translate to the user's personal preferences for the underlying atomic items and vice versa.

Covariate-dependent uncertainty quantification in simulation-based inference is crucial for high-stakes decision-making but remains challenging due to the limitations of existing methods such as conformal prediction and classical bootstrap, which struggle with covariate-specific conditioning. We propose Efficient Quantile-Regression-Based Generative Metamodeling (E-QRGMM), a novel framework that accelerates the quantile-regression-based generative metamodeling (QRGMM) approach by integrating cubic Hermite interpolation with gradient estimation. Theoretically, we show that E-QRGMM preserves the convergence rate of the original QRGMM while reducing grid complexity from $O(n^{1/2})$ to $O(n^{1/5})$ for the majority of quantile levels, thereby substantially improving computational efficiency. Empirically, E-QRGMM achieves a superior trade-off between distributional accuracy and training speed compared to both QRGMM and other advanced deep generative models on synthetic and practical datasets. Moreover, by enabling bootstrap-based construction of confidence intervals for arbitrary estimands of interest, E-QRGMM provides a practical solution for covariate-dependent uncertainty quantification.