Deep research -- producing comprehensive, citation-grounded reports by searching and synthesizing information from hundreds of live web sources -- marks an important frontier for agentic systems. To rigorously evaluate this ability, four principles are essential: tasks should be (1) user-centric, reflecting realistic information needs, (2) dynamic, requiring up-to-date information beyond parametric knowledge, (3) unambiguous, ensuring consistent interpretation across users, and (4) multi-faceted and search-intensive, requiring search over numerous web sources and in-depth analysis. Existing benchmarks fall short of these principles, often focusing on narrow domains or posing ambiguous questions that hinder fair comparison. Guided by these principles, we introduce LiveResearchBench, a benchmark of 100 expert-curated tasks spanning daily life, enterprise, and academia, each requiring extensive, dynamic, real-time web search and synthesis. Built with over 1,500 hours of human labor, LiveResearchBench provides a rigorous basis for systematic evaluation. To evaluate citation-grounded long-form reports, we introduce DeepEval, a comprehensive suite covering both content- and report-level quality, including coverage, presentation, citation accuracy and association, consistency and depth of analysis. DeepEval integrates four complementary evaluation protocols, each designed to ensure stable assessment and high agreement with human judgments. Using LiveResearchBench and DeepEval, we conduct a comprehensive evaluation of 17 frontier deep research systems, including single-agent web search, single-agent deep research, and multi-agent systems. Our analysis reveals current strengths, recurring failure modes, and key system components needed to advance reliable, insightful deep research. Our code is available at: https://github.com/SalesforceAIResearch/LiveResearchBench.

Large Language Models (LLMs) exhibit impressive capabilities but often hallucinate, confidently providing incorrect answers instead of admitting ignorance. Prior work has shown that models encode linear representations of their own knowledge and that activation steering can reduce hallucinations. These approaches, however, require real-time monitoring and intervention during inference. We introduce Contrastive Activation Steering for Amortized Learning (CASAL), an efficient algorithm that connects interpretability with amortized optimization. CASAL directly bakes the benefits of activation steering into model's weights. Once trained, LLMs answer questions they know while abstaining from answering those they do not. CASAL's light-weight design requires training only a submodule of a single transformer layer and yet reduces hallucination by 30%-40% across multiple short-form QA benchmarks. CASAL is 30x more compute-efficient and 20x more data-efficient than strong LoRA-based baselines such as SFT and DPO, boosting its practical applicability in data scarce domains. Importantly, CASAL also generalizes effectively to out-of-distribution (OOD) domains. We showcase CASAL's flexibility in mitigating hallucinations in both text-only and vision-language models. To our knowledge, CASAL is the first steering-based training method that has been shown to be effective for both dense and Mixture-of-Experts (MoE) models. CASAL represents a promising step forward for applying interpretability-inspired method for practical deployment in production systems.

A large language model can be less helpful if it exhibits output response homogenization. But whether two responses are considered homogeneous, and whether such homogenization is problematic, both depend on the task category. For instance, in objective math tasks, we often expect no variation in the final answer but anticipate variation in the problem-solving strategy. Whereas, for creative writing tasks, we may expect variation in key narrative components (e.g. plot, genre, setting, etc), beyond the vocabulary or embedding diversity produced by temperature-sampling. Previous work addressing output homogenization often fails to conceptualize diversity in a task-dependent way. We address this gap in the literature directly by making the following contributions. (1) We present a task taxonomy comprised of eight task categories that each have distinct concepts of output homogenization. (2) We introduce task-anchored functional diversity to better evaluate output homogenization. (3) We propose a task-anchored sampling technique that increases functional diversity for task categories where homogenization is undesired, while preserving it where it is desired. (4) We challenge the perceived existence of a diversity-quality trade-off by increasing functional diversity while maintaining response quality. Overall, we demonstrate how task dependence improves the evaluation and mitigation of output homogenization.

As Large Language Models (LLMs) become increasingly integrated into everyday life and information ecosystems, concerns about their implicit biases continue to persist. While prior work has primarily examined socio-demographic and left--right political dimensions, little attention has been paid to how LLMs align with broader geopolitical value systems, particularly the democracy--authoritarianism spectrum. In this paper, we propose a novel methodology to assess such alignment, combining (1) the F-scale, a psychometric tool for measuring authoritarian tendencies, (2) FavScore, a newly introduced metric for evaluating model favorability toward world leaders, and (3) role-model probing to assess which figures are cited as general role-models by LLMs. We find that LLMs generally favor democratic values and leaders, but exhibit increased favorability toward authoritarian figures when prompted in Mandarin. Further, models are found to often cite authoritarian figures as role models, even outside explicit political contexts. These results shed light on ways LLMs may reflect and potentially reinforce global political ideologies, highlighting the importance of evaluating bias beyond conventional socio-political axes. Our code is available at: https://github.com/irenestrauss/Democratic-Authoritarian-Bias-LLMs.

The sidewalk outside Majed Zeidan's grocery store in San Francisco's Mission District has stayed filled with flowers, candles, memorials and pictures since his cat was crushed under a Waymo in late October. A month later, a Waymo reportedly crushed a dog. Amid the pictures of the cat, a visitor had placed a poster that said, "save the cat, kill the car". That's when Zeidan knew Kit Kat, his bodega cat, had become the face of the simmering discontent over San Francisco's growing number of self-driving cars. Residents became increasingly comfortable riding one, costumed Halloween parade goers clambered on its rooftops and danced, and pedestrians occasionally banged its bonnet to get it to give way to them.

A therapy that would once have been considered a feat of science fiction has reversed aggressive and incurable blood cancers in some patients, doctors report. The treatment involves precisely editing the DNA in white blood cells to transform them into a cancer-fighting living drug. The first girl to be treated, whose story we reported in 2022, is still free of the disease and now plans to become a cancer scientist. Now eight more children and two adults with T-cell acute lymphoblastic leukaemia have been treated, with almost two thirds (64%) of patients in remission. T-cells are supposed to be the body's guardians - seeking out and destroying threats - but in this form of leukaemia, they grow out of control.

Activewear brand Sweaty Betty has become involved in a new dispute over advertising slogans, which a period underwear company claims were copied. Kelly Newton said Sweaty Betty's use of two taglines that were very similar to her firm Nixi Body's seemed a little off, and while she could not get them trademarked she felt Sweaty Betty was taking from other female founders. Sweaty Betty said the No ifs. Ms Newton said she was speaking out after seeing personal trainer Georgina Cox reveal Sweaty Betty had offered her a settlement over a disputed slogan . Ms Newton, who co-founded Nixi Body in 2019, said the company has advertised its leak-proof period underwear with the lines Keeping you moving through menstruation, motherhood and menopause and No leaks.

What is in the 28-point US plan for Ukraine? 'Ukraine is running out of men, money and time' Can the US get all sides to end the war? Why is Europe opposing Trump's peace plan? Firefighters evacuated residents and their pets from a nine-storey apartment building in Ukraine's Sumy region after a Russian drone strike. The strikes come as Ukrainian President Volodymyr Zelenskyy met with leaders of the UK, France and Germany in London to discuss the US peace plan.

With artificial intelligence putting productivity on hyperspeed, the painstaking but often slow nature of dealing with other countries, as well as policymaking, is also forced to speed up. But a panel at the forefront of these changes at the BRIDGE Summit in Abu Dhabi--which convenes creators, policymakers, investors, technologists, media institutions, and cultural leaders around the world to discuss the future of media--said that breaking things fast is not without consequences. "Decision makers are being asked to make decisions very quickly on the basis of information that may not be verified or verifiable," Elizabeth Churchill, a professor of Human-Computer Interaction from the Mohamed Bin Zayed University of Artificial Intelligence, told moderator Nikhil Kumar, an executive editor at TIME, which is a media partner of the BRIDGE Summit. Churchill, who held senior roles in firms like Google and Yahoo, said she returned to academia to explore transparent and "interrogable" AI tools and content that is effectively watermarked--so that decision-makers know at a glance if information is trustworthy. She said current shortfalls in information quality are "very much a design problem that sits at the surface of all of the tools that we use and in diplomacy conversations many different people are using."

Two armed men have stolen eight engravings by French artist Matisse and at least another five by Brazilian painter Cândido Portinari from a library in São Paulo. Brazilian officials say the thieves held up a security guard and an elderly couple who were visiting the library before making off with the artworks on foot. They reportedly entered the library by the main entrance at 10:00 (13:00 GMT) on Sunday, and left by the same route, heading towards the nearest metro station. The heist comes less than two months after the art world was rocked by a brazen break-in at the Louvre museum in Paris, where thieves made off with priceless jewels. The engravings stolen from Biblioteca Mário de Andrade on Sunday formed part of a joint exhibition with the São Paulo Museum of Modern Art.