Discussion | Link

Sheera Frenkel / New York Times: Anduril raised a $5B Series H led by Thrive and a16z at a $61B valuation, up from $30.5B in June 2025, taking its total funding to $6.82B, and may IPO in 2027  —  The start-up, which makes A.I.-backed weapons, was valued at $61 billion in the financing round, double what it was a year ago.

A new quantum-inspired algorithm has cracked a problem so massive that conventional supercomputers struggle to even approach it. Researchers used the method to simulate extraordinarily complex quantum materials known as quasicrystals, opening the door to powerful new quantum devices and ultra-efficient electronics. The work could help scientists design advanced topological qubits and materials for future quantum computers.

Discussion | Link

Rafe Rosner-Uddin / Financial Times: Q&A with Amazon SVP of Devices and Alexa Panos Panay on Alexa+, the company's focus on devices for the home, its Leo satellite broadband business, and more  —  Executive tasked with driving ‘beauty’ and profitability at group's device unit on hardware push that follows Alexa+

Bloomberg: How the Strait of Hormuz blockade is disrupting global chip supply chains, affecting supplies of helium, bromine, sulfur, and thinners essential to chipmaking  —  Asian chipmakers, including giants Taiwan Semiconductor Manufacturing Co. and Samsung Electronics Co., are facing disruptions …

Owen Walker / Financial Times: China's move to block Meta's Manus acquisition challenges the “Singapore washing” model, as major Chinese tech companies build significant Singapore presences  —  Beijing's move to block takeover of AI start-up with headquarters in city-state challenges sustainability of ‘Singapore washing’

This week, Nordic Compass launches, a new pan-Nordic industry alliance for resilience and competitiveness. The ambition is to move faster than the current European process and turn Nordic industrial ...

Viola Zhou / Rest of World: How the “Facebook House” in Los Altos, Mark Zuckerberg's former residence, became a hub for Chinese AI talent, who are a key part of Silicon Valley's AI boom  —  Chinese-born tech workers have fueled Silicon Valley for decades.  In the AI era, they're superstars.

Duolingo CEO Luis von Ahn said there's some work AI cannot replace. Bloomberg/Getty Images Duolingo's CEO said AI can't achieve the "level of creativity or the polish" that his top designers can. Luis von Ahn said he won't lower the quality of his team's work by pushing them to use AI unnecessarily. Duolingo has backtracked on evaluating its employees' performance based on their AI usage. Duolingo's CEO said AI still can't match the work of his company's artists and designers. On Tuesday's episode of the "Rapid Response" podcast, CEO Luis von Ahn said that his team is trying to use AI as much as possible, but "we really don't want to decrease quality." "For some things, AI is quite ready to do high-quality work. For some things, it's just not," he said. "We're not going to decrease quality just for the sake of using AI." When podcast host Bob Safian asked where AI still falls short, von Ahn pointed to design. "For example, we hire a lot of artists and designers, and our app is very high-craft when it comes to design," he said. "We're just not seeing AI get to the level of creativity or the level of polish that our top people have, by any means." Duolingo has publicly embraced AI. Last April, the company said it would evaluate employee performance partly based on AI usage. The company later walked that back, with von Ahn saying that this rule prompted employees to use AI in areas where it was unnecessary. "I don't think that was right," he said on the Rapid Response episode about the evaluation criteria. He said that while most employees benefit from using AI, there were projects or roles where it might not help. "So, making a blanket statement that we were going to evaluate employees on their usage of AI was not needed," he added. "We've removed that." AI tools have spread across creative industries, with companies increasingly using them for everything from marketing assets to product design. Some companies, like Kate Spade and Coach parent Tapestry, have said that AI is already part of their designers' workflows. Read the original article on Business Insider

Discussion | Link

Members of the French 93rd Mountain Artillery Regiment take part in live-firing exercises using Caesar mobile Howitzer systems in 2024. Leon Neal/Getty Images The EU wants to surge its arms industry, but its top diplomat says she's frustrated at its pace. The bloc is supposed to spend an extra $940 billion on defense by 2030. But there's not much information yet about who will spend that money or where it will go. The European Union's top diplomat, Kaja Kallas, said on Tuesday that she was frustrated about the bloc's pace in scaling up its defense industry, which it plans to pour an additional $940 billion into by 2030. "I share your frustration because I have this same feeling," Kallas told a reporter in Brussels who asked about the rearmament effort. "Somehow, we haven't seen the industry ramping up as we would have expected." Kallas said some European production had increased, including lines for ammunition, air defense, and critical capabilities. But the EU foreign policy chief said that, at the same time, each member state is asking for more stock. "I bring you one example that was raised today, and really needs to move faster, is the Procurement Directive. Many were raising this," she said. The Directive is an EU law that compels member states to open their contracts to defense manufacturers across the bloc, rather than rely on their domestic producers. While the law is meant to make the union's defense network fairer and help avoid redundancies, it's also been criticized as introducing too much bureaucracy. Kallas added that defense firms were concerned the EU lacked clear rules, with every country trying to order systems and ammo based on its needs. "The industry was saying also: 'You have different rules everywhere, and it's hard to operate,'" Kallas said. "Also, the standards, every member state is tweaking things a bit, so everybody has different orders, you can't really produce in line like this." While the bloc has solved some issues, "there's a lot to do," she added. The EU said in March 2025 that it hoped each member state would increase defense spending to 1.5% of its GDP by 2030. As part of that plan, the bloc said it would loosen fiscal rules to allow another $762 billion in spending over the next four years and create a $176 billion loan program for joint defense projects. The boost aims to help Europe rearm amid fears of war with Russia, but the details remain murky, including who would spend the money and where the new funds would go. Plans from Brussels largely set the direction, while each EU member state decides how much to dedicate to defense and where to allocate the money. The US, meanwhile, is poised to spend $962 billion on its military in its 2026 fiscal year, which ends on September 30. The Trump administration has said that it hopes to bump the defense budget to $1.5 trillion for the 2027 fiscal year. Read the original article on Business Insider

Walmart says it is cutting roles to reduce duplication. Brandon Bell/Getty Images Walmart is cutting or relocating 1,000 corporate jobs to address redundancies. A person familiar with the restructuring said it was not driven by AI automation. Read the full memo to employees below. Walmart is cutting or relocating about 1,000 corporate jobs to address redundancies and duplicate roles. A spokesperson for the retail giant confirmed the number of roles affected to Business Insider on Tuesday. The Wall Street Journal earlier reported the news. A person familiar with the restructuring said it was not driven by AI automation. Two executives — Suresh Kumar, the chief technology and development officer, and Daniel Danker, the company's executive vice president of AI acceleration, product, and design — announced the changes in a memo shared with employees on Tuesday. "We've made changes to simplify how the work is organized, make ownership clearer, and better align roles to the work and skills we need going forward," the memo read. Walmart laid off 1,500 corporate employees in May last year, saying it needed to "remove layers and complexity." The company employed 2.1 million people as of January. Read the full memo to employees: To: Walmart associates From: Suresh Kumar, Global CTO and Chief Development Officer Daniel Danker, EVP, AI Acceleration, Product and Design Date: May 12, 2026 RE: Evolving our tech and product teams for the future In the past year, within Global Tech and Product + Design, we've gone from organizing separately for Walmart U.S., Sam's Club, and our international markets to building in a unified way on a single, shared platform. This has enabled us to create once and scale globally, accelerate innovation, and reduce duplication. We've launched global capabilities like Marketplace across different markets more quickly and with greater impact. We've also seen places where we can operate more efficiently. In some cases, we've had different teams working on similar problems. That's why today we've made changes to simplify how the work is organized, make ownership clearer, and better align roles to the work and skills we need going forward. That includes updating some roles to better match the work being done, bringing teams together where it makes sense, and aligning some roles to key locations where related work is already happening. In a few cases, these changes also create opportunities for associates to get promoted into broader roles. These changes also involve difficult decisions. Some work has been consolidated, and some roles have been eliminated, and we know that has a real impact on people and teams. We're supporting those affected with care, including helping them explore other opportunities within Walmart where possible. Ultimately, we're designing a team that's set up to move faster, scale what works and deliver better experiences for customers and members everywhere. Today is an important step in continuing to evolve how we build and operate for the future. Read the original article on Business Insider

Jillian Kurovski was adopted from South Korea as a baby and grew up in the Midwest. Provided by Jillian Kurovski. Jillian Kurovski, a Ph.D. student, was adopted from South Korea as a baby and raised in the Midwest. After reconnecting with her birth family, she found a way back to South Korea through her work. Spending a year in Seoul near her birth family helped her better understand her identity. This as-told-to essay is based on a conversation with Jillian Kurovski, a 27-year-old Ph.D. student living in South Korea for a year. It has been edited for length and clarity. I was doing spider research in Guam when I got an email saying that my birth mother wanted to meet me. There were so many emotions, but mostly joy. I already had a three-day layover in South Korea planned on my way back to the US, and she agreed to see me while I was there. I spent a week learning the etiquette, trying on dresses, and practicing bowing because I wanted to look nice and be the perfect Korean daughter. I was born in Daegu, South Korea, and I was adopted at eight months old. I grew up in a very loving, stereotypical Midwest family in Iowa. It was a predominantly white community, and my adoption was never a secret. I always knew I was Korean, but I didn't quite have the words to express how that felt growing up. There's a moment from my childhood that still sticks with me: realizing how separated I felt from my family's Irish and Czech roots. I had an assignment in junior high to bring an artifact from our family's heritage. I remember struggling so much to think of what to bring. Do I bring something Czech or Irish, or something Korean? But I didn't know anything about Korea. I ended up printing out a picture of an Irish friendship ring. Even while I was presenting it, it just felt like a total lie because I had no connection to it. I just wanted the assignment to be over. She managed to meet her family on a three-day layover in South Korea, on the way back to the US from Guam on a research trip. Provided by Jillian Kurovski. Tracing my roots In 2018, I had started a birth search through my adoption agency, and it was kind of a spur-of-the-moment thing. At that point, I was an undergraduate researcher working with spiders, and I was scheduled to go to Guam to study them. My lab manager told me I could travel through either Hawaii or South Korea. I had never been back to my birth country before, and I chose the latter. I spent my whole three-day layover with my birth family, and it was really special, even though it was so short. It was hard to leave and go back to the US. After that, I kept thinking about how to come back to South Korea. I wanted to spend more time there and learn about my roots, so I applied for the Fulbright Presidential STEM scholarship to continue my Ph.D. research on spider reproduction, which brought me back to Seoul for a year. I arrived in July 2025 to start the program, and this time, my birth family was there to pick me up from the airport. A research scholarship gave Kurovski the chance to return to Seoul for a year. Provided by Jillian Kurovski. Over the past year, we've started to feel more like a normal family, as normal as we can be for our circumstances. I'm based in Seoul, but my family lives about an hour away in a different city, so we text about everyday things, like when I'm coming to visit or what I'm doing. Sometimes my dad will send me random YouTube videos or songs he likes. The biggest challenge has been language. My family doesn't speak much English, and I don't speak much Korean, but we get by. My dad has put in a lot of effort to learn English, and my siblings know enough to have conversations. But with my mom, I can feel the gap. I wish I could hear her tell a full story. I want to know about her life, and I want her to know mine. Finding my place in the world In some ways, I see a lot of overlap between my personal life and my research. By some kind of fate, the times I've been called to Korea have both been because of spiders. She says she sees parallels between the spiders she researches and her own life as an adoptee. Provided by Jillian Kurovski. I study an animal that's really misunderstood, and I think adoptees are often misunderstood too. People talk about both without really knowing anything about them. Spiders also exist in between — they're both predator and prey — and I think that's similar to how adoptees and multicultural people can feel, like we're not one thing or the other, but somewhere in the middle. A lot of my work also focuses on reproduction and mating behavior, so I think a lot about what it means to bring life into the world. It's made me appreciate both of my moms more. I think it's really hard to be a parent, and to be in either position is difficult. I used to struggle with people telling me I wasn't Korean or I wasn't American, and I didn't really know what that meant for my identity. Living in Korea has helped me make sense of that. Coming to South Korea and experiencing the culture firsthand and having my birth family pass it down to me for the first time was really special. At the same time, my adoptive family is very, very close. They're my biggest supporters, my biggest cheerleaders. Honestly, I think my adoptive mom was more excited for me to meet my birth family than I was. I realized I love being Korean, and I also love being American. There are things about both that I really appreciate, and I feel a lot more at peace with how I see myself now. Read the original article on Business Insider

<p><strong>Tool:</strong> <a href="https://tools.simonwillison.net/csp-allow">CSP Allow-list Experiment</a></p> <p>An experiment that shows that you can load an app in a CSP-protected sandboxed iframe (see <a href="https://simonwillison.net/2026/Apr/3/test-csp-iframe-escape/">previous note</a>) and have a custom <code>fetch()</code> that intercepts CSP errors and passes them up to the parent window... which can then prompt the user to add that domain to an allow-list and then refresh the page.</p> <p><img alt="Screenshot of a web tool titled &quot;CSP Allow-list Experiment&quot; with buttons Reset sample, Clear allow-list, Refresh preview. Left panel shows HTML source code starting with &lt;!doctype html&gt;. Right panel shows Preview with CSP header default-src 'none'; script-src 'unsafe-inline'; style-s... and heading &quot;Sandbox fetch test&quot;. A modal dialog from tools.simonwillison.net is overlaid reading: &quot;The sandbox tried to connect to: https://api.inaturalist.org Add this origin to the CSP connect-src allow-list and refresh the page?&quot; with an unchecked checkbox &quot;Don't allow tools.simonwillison.net to prompt you again&quot; and Cancel and OK buttons. Below is &quot;Messages from sandbox&quot; showing fetch-catch blocked https://api.inaturalist.org/v1/observations?per... connect-src · https://api.inaturalist.org. At the bottom left is &quot;Allowed fetch() origins&quot; with an input field containing https://api.github.com, an Add button, and a tag https://api.github.com x." src="https://static.simonwillison.net/static/2026/csp-allow.jpg" /></p> <p>I built this one with GPT-5.5 xhigh running in the Codex desktop app.</p> <p>Tags: <a href="https://simonwillison.net/tags/content-security-policy">content-security-policy</a>, <a href="https://simonwillison.net/tags/iframes">iframes</a>, <a href="https://simonwillison.net/tags/security">security</a></p>

Discussion | Link

The little pauses, “ums,” and moments when you struggle to find the right word may reveal far more about your brain than anyone realized. Researchers discovered that everyday speech patterns are closely tied to executive function — the mental system that powers memory, planning, focus, and flexible thinking. By using AI to analyze natural conversations, the team found they could predict cognitive performance with surprising accuracy, potentially opening the door to simple speech-based tools that could detect early signs of dementia long before traditional testing does.

arXiv:2605.11166v1 Announce Type: new Abstract: Political and social identities structure how people evaluate political information, a finding decades deep in political science and routinely discarded by computational tools that often produce single scores that treat a piece of text, an image, or a video as if it means the same thing to everyone. This paper shows that it does not, and that the difference is consequential. To address this problem, I develop the Perspectivist Visual Political Sentiment (PVPS) classifier, which learns from approximately 82,000 evaluations by 5,575 U.S. adults to predict how audiences defined by political and social identities will evaluate the same image. Unlike standard tools that average systematic disagreement away, PVPS preserves it, returning an evaluative profile that records who agrees, who diverges, and along which identity lines. Applied to several influential studies of visual sentiment, PVPS shows that perceived violence in protest imagery and the emotional mechanisms behind protest image engagement both change substantively once audience identity is taken into account. It follows that what a political image conveys is a moving target, and measuring it requires knowing whom it is moving.

arXiv:2605.11061v1 Announce Type: new Abstract: The evolution of visual generative models has long been constrained by fragmented architectures relying on disjoint text encoders and external VAEs. In this report, we present HiDream-O1-Image, a natively unified generative foundation model via pixel-space Diffusion Transformer, that pioneers a paradigm shift from modular architectures to an end-to-end in-context visual generation engine. By mapping raw image pixels, text tokens, and task-specific conditions into a single shared token space, HiDream-O1-Image achieves a structural unification of multimodal inputs within an Unified Transformer (UiT) architecture. This native encoding paradigm eliminates the need for separate VAEs or disjoint pre-trained text encoders, allowing the model to treat diverse generation and editing tasks as a consistent in-context reasoning process. Extensive experiments show that HiDream-O1-Image excels across various generation tasks, including text-to-image generation, instruction-based editing, and subject-driven personalization. Notably, with only 8B parameters, HiDream-O1-Image (8B) achieves performance parity with or even surpasses established state-of-the-art models with significantly larger parameters (e.g., 27B Qwen-Image). Crucially, to validate the immense scalability of this paradigm, we successfully scale the architecture up to over 200B parameters. Experimental results demonstrate that this massive-scale version HiDream-O1-Image-Pro (200B+) unlocks unprecedented generative capabilities and superior performance, establishing new state-of-the-art benchmarks. Ultimately, HiDream-O1-Image highlights the immense potential of natively unified architectures and charts a highly scalable path toward next-generation multimodal AI.

arXiv:2605.10981v1 Announce Type: new Abstract: Reference-free preference optimization has emerged as an efficient alternative to reinforcement learning from human feedback, with Simple Preference Optimization(SimPO) demonstrating strong performance by eliminating the explicit reference model through a simple objective. However, the joint tuning of the hyperparameters $\beta$ and $\gamma$ in SimPO remains a central challenge. We argue that this difficulty arises because the margin formulation in SimPO is not easily interpretable across datasets with different reward gap structures. To better understand this issue, we conduct a comprehensive analysis of SimPO and find that $\beta$ implicitly controls sample filtering, while the effect of $\gamma$ depends on the reward gap structure of the dataset. Motivated by these observations, we propose $\xi$-DPO: Direct preference optimization via ratio reward margin. We first reformulate the preference objective through an equivalent transformation, changing the optimization target from maximizing the likelihood of reward gaps to minimizing the distance between reward gaps and optimal margins. Then, we redefine the reward in a ratio form between the chosen and rejected, which effectively cancels the effect of $\beta$ and yields a bounded and interpretable margin. This margin is called the ratio reward margin and is denoted by $\xi$. Unlike the margin $\gamma$ in SimPO, $\xi$ explicitly represents the desired relative separation between chosen and rejected responses and can be determined from the initial reward gap distribution, avoiding repeated trial-and-error tuning. ....