Chances are, you've heard of the term "large language models," or LLMs, when people are talking about generative AI. But they aren't quite synonymous with the brand-name chatbots like ChatGPT, Google Gemini, Microsoft Copilot, Meta AI and Anthropic's Claude.These AI chatbots can produce impressive results, but they don't actually understand the meaning of words the way we do. Instead, they're the interface we use to interact with large language models. These underlying technologies are trained to recognize how words are used and which words frequently appear together, so they can predict future words, sentences or paragraphs. Understanding how LLMs work is key to understanding how AI works. And as AI becomes increasingly common in our daily online experiences, that's something you ought to know.This is everything you need to know about LLMs and what they have to do with AI.What is a language model?You can think of a language model as a soothsayer for words."A language model is something that tries to predict what language looks like that humans produce," said Mark Riedl, professor in the Georgia Tech School of Interactive Computing and associate director of the Georgia Tech Machine Learning Center. "What makes something a language model is whether it can predict future words given previous words."This is the basis of autocomplete functionality when you're texting, as well as of AI chatbots.What is a large language model?A large language model contains vast amounts of words from a wide array of sources. These models are measured in what is known as "parameters."So, what's a parameter?Well, LLMs use neural networks, which are machine learning models that take an input and perform mathematical calculations to produce an output. The number of variables in these computations are parameters. A large language model can have 1 billion parameters or more."We know that they're large when they produce a full paragraph of coherent fluid text," Riedl said.How do large language models learn?LLMs learn via a core AI process called deep learning."It's a lot like when you teach a child -- you show a lot of examples," said Jason Alan Snyder, global CTO of ad agency Momentum Worldwide.In other words, you feed the LLM a library of content (what's known as training data) such as books, articles, code and social media posts to help it understand how words are used in different contexts, and even the more subtle nuances of language. The data collection and training practices of AI companies are the subject of some controversy and some lawsuits. Publishers like The New York Times, artists and other content catalog owners are alleging tech companies have used their copyrighted material without the necessary permissions.(Disclosure: Ziff Davis, CNET's parent company, in April filed a lawsuit against OpenAI, alleging it infringed on Ziff Davis copyrights in training and operating its AI systems.)AI models digest far more than a person could ever read in their lifetime -- something on the order of trillions of tokens. Tokens help AI models break down and process text. You can think of an AI model as a reader who needs help. The model breaks down a sentence into smaller pieces, or tokens -- which are equivalent to four characters in English, or about three-quarters of a word -- so it can understand each piece and then the overall meaning.From there, the LLM can analyze how words connect and determine which words often appear together."It's like building this giant map of word relationships," Snyder said. "And then it starts to be able to do this really fun, cool thing, and it predicts what the next word is … and it compares the prediction to the actual word in the data and adjusts the internal map based on its accuracy."This prediction and adjustment happens billions of times, so the LLM is constantly refining its understanding of language and getting better at identifying patterns and predicting future words. It can even learn concepts and facts from the data to answer questions, generate creative text formats and translate languages. But they don't understand the meaning of words like we do -- all they know are the statistical relationships.LLMs also learn to improve their responses through reinforcement learning from human feedback."You get a judgment or a preference from humans on which response was better given the input that it was given," said Maarten Sap, assistant professor at the Language Technologies Institute at Carnegie Mellon University. "And then you can teach the model to improve its responses." LLMs are good at handling some tasks but not others.

05/27 — 2025 by abduzeedo Explore Carlos el Rojo's "Visual Games" — a masterclass in illustration, wit, and conceptual design. Discover his unique approach to visual storytelling. Hey, design enthusiasts! Today, let's dive into the captivating world of Carlos el Rojo, an illustrator whose work truly embodies the spirit of playful yet profound visual communication. His series, aptly titled "Visual Games," offers a fresh perspective on how ideas can take shape with minimal fuss but maximum impact. Carlos el Rojo, a celebrated illustrator, approaches his creations as "small visual ideas, quick to execute, but with a twist". These aren't your typical commissioned pieces. Instead, they are a testament to his dedication to "training the eye and saying something meaningful with as little as possible". This philosophy resonates deeply with anyone passionate about concise and powerful illustration. The Spark of Inspiration The journey into "Visual Games" began for Carlos after a transformative workshop with Javier Jaén. This experience, as he describes, "helped me unlock something". It fostered a habit of embracing ideas without overthinking, allowing for a fluid, organic creative process. This approach is something many designers can relate to—the freedom to just go with it when inspiration strikes. His work often carries layers of meaning. Sometimes it is ironic, sometimes poetic, and occasionally, it might even be a little uncomfortable. Yet, each piece shares a common thread: "they're made with curiosity and without pretense". This genuine curiosity is what makes his illustration so compelling. Each "game" is a lesson in visual identity and concise storytelling. They are not just pretty pictures; they are small puzzles, contradictions, or metaphors that challenge the viewer to think beyond the obvious. This approach to illustration is a masterclass in making every element count. The Power of Play Carlos el Rojo's "Visual Games" serve as a potent reminder that design, at its core, is about problem-solving and communication. By embracing curiosity and letting ideas flow, he creates work that is both engaging and thought-provoking. His illustrations are a testament to the power of a simple idea executed with precision and a touch of genius. For more inspiration and to explore his full portfolio, be sure to visit Carlos el Rojo's official website: carloselrojo.com and his Instagram. Art and illustration artifacts Tags art illustration

By Ellyn Lapointe Published May 31, 2025 | Comments (0) | Researchers recovered this face-shaped rock from the San Lázaro rock-shelter in central Spain in 2022 © Álvarez-Alonso et al While digging inside a cave in the Spanish city of Segovia, archaeologists uncovered an unusual rock. The hand-sized stone naturally resembled an elongated face, and featured a spot of red pigment made from ochre right on the tip of what may be considered its nose.  “We were all thinking the same thing and looking at each other because of its shape: we were all thinking, ‘This looks like a face,’” David Álvarez Alonso, an archaeologist at Complutense University in Madrid who was part of the dig, told The Guardian. Álvarez Alonso and his colleagues spent the next three years studying this bizarre rock. The researchers posit that 43,000 years ago, a Neanderthal dipped their finger in ochre and pressed it onto the stone’s central ridge—leaving behind what is now considered to be the world’s oldest complete human fingerprint. It’s an intriguing finding that could have significant implications, but some experts would like to see more evidence to support this hypothesis.  The team published its findings in the journal Archaeological and Anthropological Sciences on Saturday, May 24. In the paper, the archaeologists state that the “strategic position” of the dot suggests it is evidence of Neanderthals’ “symbolic behavior.” In other words, it’s a piece of art that “could represent one of the earliest human face symbolizations in prehistory.” “The fact that the [rock] was selected because of its appearance and then marked with ochre shows that there was a human mind capable of symbolizing, imagining, idealizing and projecting his or her thoughts on an object,” the researchers write.  Whether Neanderthals were capable of making art is a subject of ongoing debate, co-author María de Andrés-Herrero, a professor of prehistory at Complutense University, told the BBC. But over the past decade, a growing body of evidence has led many experts to believe that artistic expression emerged earlier in human evolution than previously thought.  The authors of this new study think their stone adds to this evidence. To reach this conclusion, they first needed more data to support the idea that this ancient artist had actually experienced pareidolia: seeing a face in an inanimate object. To that end, they generated a 3D model of the stone’s surface and measured the distances between its features, finding that the red dot—or nose—was placed such that it accurately resembled an actual nose on a human face.  Then, the researchers enlisted the help of geologists to characterize the red dot, confirming that it was made with ochre. Forensic police experts then used multispectral analysis—a technique that can reveal details invisible to the naked eye—to confirm that the red dot had been applied with a fingertip. Their analysis uncovered a fingerprint that could have belonged to an adult male Neanderthal inside the dot.  “Once we had that and all the other pieces, context and information, we advanced the theory that this could be a pareidolia, which then led to a human intervention in the form of the red dot,” Álvarez Alonso told The Guardian. “Without that red dot, you can’t make any claims about the object.” But Gilliane Monnier, a professor of anthropology at the University of Minnesota who studies Neanderthal behavior, is not totally convinced by the researchers’ findings. “The fact that there are these natural depressions—and that we can measure the distance between them and argue that it’s a face—that’s all well and good,” Monnier, who wasn’t involved in the study, told Gizmodo. “But that doesn’t give us any indication that the Neanderthals who [occupied this cave] saw a face in that [rock].”  What’s more, she is skeptical of the researchers’ claim that the red dot was actually made with a human fingertip. It’s possible, she said, that the coloring and fingerprint-like ridges formed naturally. “I would be interested in seeing an explanation by a geologist—someone trained in geology—saying the likelihood of this forming by natural, geological or geomicrobial processes is a very low likelihood,” Monnier said.  The researchers, too, acknowledge that “it is unlikely that all doubts surrounding this hypothesis can be fully dispelled,” and state that the pareidolia hypothesis should not be seen as a definitive claim, but rather a possible explanation for this object based on the evidence. So it’s hard to say whether this study clarifies or complicates our understanding of how the human mind evolved the ability to create art. The face-shaped rock is an intriguing piece of the puzzle, but more research is needed to figure out where it fits. Daily Newsletter You May Also Like By Margherita Bassi Published May 28, 2025 By Margherita Bassi Published May 27, 2025 By Margherita Bassi Published May 25, 2025 By Natalia Mesa Published May 13, 2025 By Margherita Bassi Published May 8, 2025 By Margherita Bassi Published April 23, 2025

Thermalright is a Menace: Dozens of New Coolers, New Case, 17-Blade Fan, & Mini PCs, ft. CEOMay 30, 2025Last Updated: 2025-05-30We take a look at Thermalright’s crazy amount of air coolers, closed-loop liquid coolers, fans, and examine the company’s first caseThe HighlightsThermalright continues to overwhelm the CPU cooler market with a crazy amount of coolers at cheap pricesThermalright showed off its first case, the TR M10 MATXThermalright is experimenting with fans of all kinds, ranging from high-end all-LCP and metal fans to more modest PPT onesThe company also unveiled a number of new closed-loop liquid coolersTable of ContentsAutoTOCGrab a GN Soldering & Project Mat for a high-quality work surface with extreme heat resistance. These purchases directly fund our operation, including our build-out of the hemi-anechoic chamber for our acoustic testing! (or consider a direct donation or a Patreon contribution!)IntroWe visited Thermalright’s booth at Computex 2025 and the company handed us a big booklet showing everything the company was showing at the show. It’s impossible for us to remember the names of everything the company was showing at the event as there just too much of it, but we'll try to highlight the most interesting products.Editor's note: This was originally published on May 20, 2025 as a video. This content has been adapted to written format for this article and is unchanged from the original publication.CreditsHostSteve BurkeCamera, Video EditingMike GaglioneVitalii MakhnovetsWriting, Web EditingJimmy ThangThermalright Air CoolersThe biggest thing that Thermalright is doing when it comes to coolers is mostly in digital displays and screens. The company showed off many CPU air coolers at Computex and only a few don’t have them. Frost TowerThe company’s Frost Tower comes with a 140mm fan in the middle coupled with a 120mm front fan. The middle fan is 27mm thick and the front fan is a standard 25mm. Thermalright is targeting about $50 for the Frost Tower. The company is using LCP blades, which is very expensive, and a 30% fiberglass reinforced PBT for the frame, which is a balancing mechanism for cost. It uses a 6-pull fan and its fin-stack is soldered, which is higher quality than press-fit.Royal LordThe company’s Royal Lord cooler is supposed to be a $43 cooler and it uses 2 fans. Thermalright says it’s an extra $10 if you want to add an extra LCD to it. Like the Frost Tower, its fin stack is also soldered. It uses 7 heat pipes with a 30% fiberglass reinforced PBT for its fans, which are both 2x25mm. Thermalright FansR-Series Fans The company will release its R5 and R9 fans. The number denotes how many blades the fans will have. Thermalright says the R5 fans are designed to push air through radiators. The R9 fan has 9 blades and is a 28mm thick fan. The company is moving towards LCP on a lot of its fans. X12 FanThermalright showed off a fan with 17 blades that’s completely made out of metal. It’s got a zinc alloy frame and aluminum blade set. It’s just a prototype for now and costs the company $100 to make, which tells you a little about how much it might cost for the company to mass-produce it. The unit we saw was fully CNCed. 17 blade fans are very abnormal. The company says it would mostly be useful in mini PCs. The fan uses an all-metal design to get the blade-tip clearance to functionally be 0. The fan also has an inner ring to it. Thermalright Stream Vision Liquid Cooler Grab a GN15 Large Anti-Static Modmat to celebrate our 15th Anniversary and for a high-quality PC building work surface. The Modmat features useful PC building diagrams and is anti-static conductive. Purchases directly fund our work! (or consider a direct donation or a Patreon contribution!)Thermalright showed off numerous liquid coolers, but we’re only going to focus on one. Thermalright’s Stream Vision is supposed to be $100. The liquid cooler comes with a screen on it and there’s also a fan that cools the memory and VRM. One of our suggestions to Thermalright is to angle the grill on the sides of this fan away from the mounting brackets. The company is trying to compete in the liquid market more. Rainbow Vision and Wonder VisionThe company is also competing with Tryx’s panoramic cooler. Thermalright has an interchangeable screen with one of the 2 options providing a full separate cold plate. They are 2400x1800 resolution from what the company told us and leverage a 60Hz panel. One of the options is a 6.67-inch OLED display. Thermalright says that both options should be about $200. TRM10 CasesThermalright also showed off a case that it’s been working on, the TRM10. The company hasn’t made computer cases before. It’s one of the few products that it’s not making in its factory. Thermalright’s plan for cases is to do what they’re doing in the cooler industry, which is to be extremely competitive when it comes to pricing. The cheapest version of the TRM10 will be $45 and the more expensive one will be $65. The more expensive model will come with an LCD display on the side and a digital display on the front to display numbers like the time. The cheaper one, on the other hand, will most likely just come with a steel plate instead.  The mATX case is mostly steel with a bunch of perforations on the top and glass for its sides. The top of the case also has plastic that is made to look like brushed aluminum. Inside the chassis, there’s a good amount of depth for the cable management and it has passthrough for the cables at a side angle with the exception of the EPS12V, which is on the top. Everything else about the case is pretty standard. Looking inside the case, there’s a lot of perforation, like at the top and bottom of the hard drive cages, which is a good thing to see. The top of the power supply shroud is also heavily perforated along with the bottom of the case. There’s about as many holes as you can get in the case.  The case is going for airflow with its panel design, with the exception of the front, but the side makes up for some airflow. The steel in the middle of this panel adds some rigidity and is also probably a play on Thermalright’s logo.   Thermalright Mini PCsWith its mini PCs, Thermalright is trying to compete with Minisforum and the company tells us that the mini PCs we saw will be liquid cooled. This theoretically means it should have better thermals and be a little quieter compared to competing systems if it’s done well. Pricing is still TBD but Thermalright says they will be cheaper or equal to competing mini PCs, which we take as Minisforum. The company started running a benchmark on one of its mini PCs with a 100% load and we saw the CPU, which is an AMD 395 SOC, temps being 61 degrees C, though that may ramp up as it hits steady state. Thermalright showed off 3 models.One of the mini PCs at Computex was just there for show. The unit in the image above is fully CNCed. It looks super nice with its red accents, but it’s not something the company is planning to sell unless there’s a lot of demand for it.  Thermalright Tank PCLeveraging its factory, Thermalright wanted to build something cool and took 3 months to make a computer that looked like a tank. The company told us it’s composed of around 450 pieces.Thermalright CEO Interview Visit our Patreon page to contribute a few dollars toward this website's operation (or consider a direct donation or buying something from our GN Store!) Additionally, when you purchase through links to retailers on our site, we may earn a small affiliate commission.We also conducted a brief interview with Thermalright’s CEO. To check that out, make sure to watch that portion in our Computex video.

The $7.6 billion Hyundai Motor Group Metaplant America, or HMGMA, is one of the newest and most technologically advanced car factories in the world.The plant, located near Savannah, Georgia, opened its doors in March and will be a key production facility for Hyundai's EVs and PHEVs, as well as those belonging to its Genesis luxury brand and sister company Kia.In a recent interview with Business Insider, Genesis North America COO Tedros Mengiste cited the investment as an example of Hyundai's track record for "visionary and strategic, and long-term thinking."I recently took a behind-the-scenes tour of Hyundai's new megafactory packed with autonomous robots and state-of-the-art tech. The Hyundai Metaplant is situated on a 3,000-acre campus in the south Georgia town of Ellabell. Hyundai's Metaplant America. Hyundai Located just 20 miles from the Port of Savannah, one of the busiest in the US, the plant not only gives Hyundai much-needed manufacturing capacity in the US to avoid import tariffs, but it also affords the company the flexibility to export vehicles abroad.It also gives Hyundai the production footprint to compete against rivals like Tesla, GM, and Rivian, which is also building a new factory in Georgia. Driving up to the factory, it's easy to be wowed by the sheer scale of the sprawling complex. The entryway to the Hyundai Motor Group Metaplant America campus in Ellabell, Georgia. Benjamin Zhang/Business Insider It's Hyundai Group's second car factory in the state. The company also operates a $3.2 billion, 2,200-acre facility in West Point, Georgia, that builds Kia EV and ICE SUVs. I drove to the factory in a new 2026 Hyundai Ioniq 9 EV SUV, which is one of the vehicles assembled at the Metaplant. Hyundai Ioniq 9 EVs are parked in front of the lobby at the Hyundai Motor Group Metaplant America in Georgia. Hyundai The only other model assembled at the plant is the Hyundai Ioniq 5 EV. My tour began in the plant's modern main lobby. The Metaplant lobby is modern and pleasant. Benjamin Zhang/Business Insider Hyundai broke ground on the facility in the fall of 2022 and took just two years to complete construction on the main production buildings. The Metaplant site consists of 11 buildings totalling 7.5 million square feet of space. A map of the Hyundai Motor Group Metaplant America in Georgia. Benjamin Zhang/Business Insider The Metaplant is a marvel of vertical integration, with the goal of having as many key components, ranging from battery packs to seats, made on-site. Here's a Hyundai XCIENT hydrogen fuel cell semi truck used to transport parts and supplies to the factory. A Hyundai XCIENT hydrogen fuel cell truck. Benjamin Zhang/Business Insider It's one of 21 emission-free XCIENT trucks deployed around the Metaplant site. The production process starts in the stamping shop, where sheet metal is cut and stamped into parts that will make up the frame of the car. The stamping facility. Benjamin Zhang/Business Insider The sheet metal is supplied by the on-site Hyundai Steel facility. Stamped parts are transported by automated guided vehicles, or AGVs. Autonomous robots are transporting stamped metal parts. Benjamin Zhang/Business Insider The plant employs almost 300 AGVs to shuttle everything from spare parts to partially assembled cars. The stamped metal panels are then stored in these massive racks. Racks full of stamped metal sections of Ioniq 5 and Ioniq 9 EVs. Benjamin Zhang/Business Insider The Metaplant was originally expected to produce up to 300,000 electrified vehicles annually. However, Hyundai announced at the plant's grand opening in March that its capacity will be expanded to 500,000 units in the coming years as part of a new $21 billion investment in US manufacturing. Here are parts of the Ioniq 9, Hyundai's new flagship three-row EV SUV. Parts of the Hyundai Ioniq 9 EV at the Hyundai Motor Group Metaplant America in Georgia. Benjamin Zhang/Business Insider The plant is expected to start production of its first Kia model next year. The next part of the tour is the welding shop. Ioniq 5 EVs at the welding facility at the Hyundai Motor Group Metaplant America in Georgia. Benjamin Zhang/Business Insider Here, the stamped metal pieces are welded together by robot to form the body of the vehicle. The work done by the welding robots is then inspected by the plant's human employees known as Meta Pros. The Hyundai Ioniq 5 and Ioniq 9 EVs are going through quality inspections in the welding shop. Benjamin Zhang/Business Insider The Metplant employees more than 1,300 Meta Pros, nearly 90% of whom were hired locally. There are employee meeting and break areas located along the inspection and assembly areas. Employee break and meeting area at the welding shop. Benjamin Zhang/Business Insider An employee cafeteria with remote ordering capability is located in the main assembly building. In addition to human eyes, the vehicles are also inspected by a pair of Boston Dynamics robot dogs called Spot. Boston Dynamics robot dogs inspecting Hyundai Ioniq 5 EVs. Benjamin Zhang/Business Insider In 2021, Hyundai acquired an 80% stake in Boston Dynamics in a deal that valued the company at $1.1 billion. After the inspections are complete, a robot loads the partially assembled vehicles onto a conveyor system. Ioniq 5 EVs are about to be lifted onto the conveyor belt to the paint shop. Benjamin Zhang/Business Insider Next stop, the paint shop. Unfortunately, my tour did not get access to the paint shop due to concerns that outside visitors may compromise the quality of the paint application. Hyundai EV bodies are moving from the paint shop to the assembly facility at the Hyundai Motor Group Metaplant America in Georgia. Benjamin Zhang/Business Insider After receiving a fresh coat of paint, the vehicles travel through a bridge to the assembly building. Here, the painted bodies are married with their battery packs and skateboard chassis. An Ioniq 5 on the assembly line. Benjamin Zhang/Business Insider Hyundai Mobis produces the skateboard chassis in a building next door to the general assembly facility. The Metaplant's on-site battery factory, operated in a joint venture with LG, is expected to come online next year. The plant currently sources its batteries from Hyundai's other facilities, including one in North Georgia that's a joint venture with SK. The vehicles' interiors are then assembled by hand. The Metaplant assembly line, where human workers are joining in. Benjamin Zhang/Business Insider The further along the production process, the more you see human workers on the assembly line. Partially assembled EVs are shuttled through from area to area by the automated robots. Ioniq 5 EVs at the Hyundai Motor Group Metaplant America in Georgia. Benjamin Zhang/Business Insider The entire facility was immaculately clean, quiet, and felt beautifully choreographed. Assembled vehicles are loaded onto different AGVs that navigate the facility by reading the QR codes embedded into the floor. Hyundai Ioniq 5 EVs after soak testing at the Hyundai Motor Group Metaplant America in Georgia. Benjamin Zhang/Business Insider These AGVs shuttle the vehicles through the plant's various quality control tests. At the end of the assembly line, completed EVs are put through their paces at the on-site test track before being sent to the vehicle preparation center, or VPC, to get them ready for shipping. Completed Hyundai EVs are ready for a dealer's lot. Benjamin Zhang/Business Insider Vehicles destined for dealerships in the region are put on trucks, while those traveling more than 500 miles are shipped by rail at the Metplant's on-site train terminal.

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Alibaba Group has introduced QwenLong-L1, a new framework that enables large language models (LLMs) to reason over extremely long inputs. This development could unlock a new wave of enterprise applications that require models to understand and draw insights from extensive documents such as detailed corporate filings, lengthy financial statements, or complex legal contracts. The challenge of long-form reasoning for AI Recent advances in large reasoning models (LRMs), particularly through reinforcement learning (RL), have significantly improved their problem-solving capabilities. Research shows that when trained with RL fine-tuning, LRMs acquire skills similar to human “slow thinking,” where they develop sophisticated strategies to tackle complex tasks. However, these improvements are primarily seen when models work with relatively short pieces of text, typically around 4,000 tokens. The ability of these models to scale their reasoning to much longer contexts (e.g., 120,000 tokens) remains a major challenge. Such long-form reasoning requires a robust understanding of the entire context and the ability to perform multi-step analysis. “This limitation poses a significant barrier to practical applications requiring interaction with external knowledge, such as deep research, where LRMs must collect and process information from knowledge-intensive environments,” the developers of QwenLong-L1 write in their paper. The researchers formalize these challenges into the concept of “long-context reasoning RL.” Unlike short-context reasoning, which often relies on knowledge already stored within the model, long-context reasoning RL requires models to retrieve and ground relevant information from lengthy inputs accurately. Only then can they generate chains of reasoning based on this incorporated information.  Training models for this through RL is tricky and often results in inefficient learning and unstable optimization processes. Models struggle to converge on good solutions or lose their ability to explore diverse reasoning paths. QwenLong-L1: A multi-stage approach QwenLong-L1 is a reinforcement learning framework designed to help LRMs transition from proficiency with short texts to robust generalization across long contexts. The framework enhances existing short-context LRMs through a carefully structured, multi-stage process: Warm-up Supervised Fine-Tuning (SFT): The model first undergoes an SFT phase, where it is trained on examples of long-context reasoning. This stage establishes a solid foundation, enabling the model to ground information accurately from long inputs. It helps develop fundamental capabilities in understanding context, generating logical reasoning chains, and extracting answers. Curriculum-Guided Phased RL: At this stage, the model is trained through multiple phases, with the target length of the input documents gradually increasing. This systematic, step-by-step approach helps the model stably adapt its reasoning strategies from shorter to progressively longer contexts. It avoids the instability often seen when models are abruptly trained on very long texts. Difficulty-Aware Retrospective Sampling: The final training stage incorporates challenging examples from the preceding training phases, ensuring the model continues to learn from the hardest problems. This prioritizes difficult instances and encourages the model to explore more diverse and complex reasoning paths. QwenLong-L1 process Source: arXiv Beyond this structured training, QwenLong-L1 also uses a distinct reward system. While training for short-context reasoning tasks often relies on strict rule-based rewards (e.g., a correct answer in a math problem), QwenLong-L1 employs a hybrid reward mechanism. This combines rule-based verification, which ensures precision by checking for strict adherence to correctness criteria, with an “LLM-as-a-judge.” This judge model compares the semanticity of the generated answer with the ground truth, allowing for more flexibility and better handling of the diverse ways correct answers can be expressed when dealing with long, nuanced documents. Putting QwenLong-L1 to the test The Alibaba team evaluated QwenLong-L1 using document question-answering (DocQA) as the primary task. This scenario is highly relevant to enterprise needs, where AI must understand dense documents to answer complex questions.  Experimental results across seven long-context DocQA benchmarks showed QwenLong-L1’s capabilities. Notably, the QWENLONG-L1-32B model (based on DeepSeek-R1-Distill-Qwen-32B) achieved performance comparable to Anthropic’s Claude-3.7 Sonnet Thinking, and outperformed models like OpenAI’s o3-mini and Qwen3-235B-A22B. The smaller QWENLONG-L1-14B model also outperformed Google’s Gemini 2.0 Flash Thinking and Qwen3-32B.  Source: arXiv An important finding relevant to real-world applications is how RL training results in the model developing specialized long-context reasoning behaviors. The paper notes that models trained with QwenLong-L1 become better at “grounding” (linking answers to specific parts of a document), “subgoal setting” (breaking down complex questions), “backtracking” (recognizing and correcting their own mistakes mid-reasoning), and “verification” (double-checking their answers). For instance, while a base model might get sidetracked by irrelevant details in a financial document or get stuck in a loop of over-analyzing unrelated information, the QwenLong-L1 trained model demonstrated an ability to engage in effective self-reflection. It could successfully filter out these distractor details, backtrack from incorrect paths, and arrive at the correct answer. Techniques like QwenLong-L1 could significantly expand the utility of AI in the enterprise. Potential applications include legal tech (analyzing thousands of pages of legal documents), finance (deep research on annual reports and financial filings for risk assessment or investment opportunities) and customer service (analyzing long customer interaction histories to provide more informed support). The researchers have released the code for the QwenLong-L1 recipe and the weights for the trained models. Daily insights on business use cases with VB Daily If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI. Read our Privacy Policy Thanks for subscribing. Check out more VB newsletters here. An error occured.

Small sectional sofas (which are a thing, trust us) are just the right seating solution for those with cramped spaces who recognize that it’s not humanly possible to get cozy on a loveseat. “A sectional can give you that maxed-out seating and help the room feel like a cozy nook, not a cramped back office,” says designer Caroline Burke of Anna Burke Interiors.Though sectionals in all their long configurations can feel counterintuitive for small spaces, designer Ashley Darryl says the opposite is often true. “Sometimes using several small pieces can make a room feel cluttered,” says of the sofa-chair-table cluster that’s become de rigueur in living rooms. Whether your goal is to fit the whole family for movie night or just successfully stretch out completely solo, the right-size sectional can transform a cramped room in unexpected ways.“Typically, sectionals can double your seating,” notes AD100 interior designer Victoria Hagan, and author of Live Now. Burke agrees, adding that choosing a piece with a low back and arms can “help a small room not feel engulfed by a piece of furniture.”To help with your seating journey, we pulled together our favorite sectionals for small spaces—from bouclé beauties to rawhide leather configurations. Read up on some common seating FAQs below, answered by designers, as well.Our Top Picks for the Best Small Sectional SofasBest Leather Option: Castlery Jonathan Leather Side Chaise Sofa, $4,547Best Upholstered Sofa: Vesgantti Modular Sectional Sofa, $700 $550Best Sleeper Sofa: Pottery Barn Shasta Reversible Chaise Sleeper Sectional, $2,599For Sculptural Seating: AllModern Anson Modular Sectional, $3,098 $2,673For Iconic Design: Herman Miller Luva Modular Sectional, $8,430For a Modular Arrangement: Burrow Range 3-Piece Sectional Lounger, $1299 $979In This ArticleBest Leather OptionCastlery Jonathan Leather Side Chaise SectionalSpecsDimensions: 27.6"H x 84.3"W x 94.1"D (long side) and 39.4"D (short side)Seat Depth: 24.4"Materials: Top grain leather, veneer lumber, and plywoodCare: Wipe clean with dry clothAD commerce director Rachel Fletcher has adored this modular sectional since 2022, reviewing it as “modern, minimalist, and tasteful,” pet-friendly, and a powerful design moment for your living room. This modular sectional includes a versatile chaise lounge so you can expand your sectional or split it up when you move.Best Budget SofaWade Logan Fultonham 6-Piece Upholstered SectionalSpecsDimensions: 27.6"H x 84.3"W x 94.1"D (long side) and 39.4"D (short side)Seat Depth: 24.4"Materials: Top grain leather, veneer lumber, and plywoodCare: Wipe clean with dry clothOkay, yes, six pieces seem like a lot for a small sectional sofa. But keep in mind that you can mix-and-match elements, turning some into cozy reading chairs and connecting others into larger lounge spaces. We love the taller back cushions on this blocky option which ensure that you aren’t slouching your way through your next movie marathon.Best Upholstered SofaVesgantti Modular Sectional SofaSpecsDimensions: 60"D x 108"W x 35.5"HSeat Depth: 26"Materials: Wood, foam fill, chenille upholsteryCare: Wipe clean with dry clothReviews of this Amazon sectional tout the benefits of durable-feeling upholstery and easy assembly. Sure, 108 inches may seem large for small spaces, but that’s only if you format this sofa as a three-seater. A modular makeup ensures that you can move each piece around to best suit your living room. As an added bonus, this sofa comes ready with two sets of legs at varying heights, so you can decide how low to the ground you want to be.For Sculptural SeatingAllModern Anson Modular SectionalSpecsDimensions: 32.5''H x 100.75''W x 61.25''DSeat Depth: Not listedMaterials: Wood, foam, synthetic fiber fill, polyester upholsteryCare: Spot clean with dry, solvent-based cleanerWith a curved, high back that draws you in like a hug, and slim lumbar pillows for added back support, All Modern’s Anson sectional takes the best elements of sculptural furniture and makes them compatible with small space living. Available in three upholstery colors and the option of chaise sides, this sectional sofa is an ideal happy medium between sink-in cozy texture and statement-worthy design.Best Sleeper SofaPottery Barn Shasta Reversible Chaise Sleeper SectionalSpecsDimensions: 83.5"W x 59"D x 35"HSeat Depth: 29.5"Materials: Solid rubberwood frame, polyester-wrapped cushions, innerspring mattress,Care: Blot spills immediately with a clean, colorfast towel or sponge, vacuum cushions regularlyHosting guests in a small space requires some compromise, but any consummate host knows that good hospitality doesn’t come in the form of a saggy air mattress. Instead, this sleek sofa folds out into a queen-sized bed, accommodating your guests comfortably without taking up too much additional space. All-white can be dicey for any highly trafficked piece of furniture, but the sofa also comes in a range of hardier fabrics like basketweave, tweed, and velvet in all sorts of rich colors. Buyers even have the option to add storage in the chaise.Best Velvet OptionArticle Abisko Velvet SectionalSpecsDimensions: 28"H x 94.5"W x 63.5"DSeat Depth: 25"Materials: Frame: kiln-dried solid larch, plywood, brushed aluminum; Filling: high-density foam, polyester fiber; Fabric: 100% polyesterCare: Blot stains with a dry clothThe best-selling Sven sectional from Article is made even more beautiful by colorful, performance velvet upholstery. Suitable for those with little ones and pets running around, the fabric is non-absorptive, meaning spills and crumbs won’t stay embedded in its weave. Two orientations—chaise and L-shaped—allow you to Tetris its silhouette into your space as you see fit.For Cloudlike UpholsteryWest Elm Harmony Modular Chaise SectionalSpecsDimensions: 86"W x 62"D x 37"HSeat Depth: 23"Materials: Engineered hardwood frame, fiber-wrapped high-density polyurethane foam, high-gauge sinuous springs, upholstery (linen, basketweave, velvet, chenille, and twill options)Care: Blot spills immediately with a clean, colorfast towel; spot clean with a damp cloth or spongeWith a seat depth of 23 inches, this is the ideal small sectional for sinking in and getting cozy. A multitude of pillows allows for added comfort, while the chaise seat lets at least one person on the end stretch their legs out. Rather than simply make this sofa smaller or more narrow, the entire size has also been scaled down to look proportional without sacrificing style in smaller living rooms.For a Thoroughly Modern SilhouetteDesign Within Reach Pastille Sectional ChaiseSpecsDimensions: 57"D x 80"W x 32"HSeat Depth: 23"Materials: Steel-reinforced plywood frame, high-density, high-resiliency foam cushions, memory foam toppers, fabric or leather upholsteryCare: Spot cleanThe Pastille Sectional with an included chaise is an appropriate ode to the British candy, combining clean lines with cheerful colorways. Designed by Hlynur Atlason, this lounge sofa features a low seat (approximately 18 inches high), for not only cramped quarters, but low-ceilinged ones, too. And thanks to the lightness of the design, this sofa will also never look too heavy or oversized for your living room.For an Artpiece SofaHerman Miller Luva Modular SectionalSpecsDimensions: 61.5"D x 78.25"W x 40.75"HSeat Depth: 51.5"Materials: Steel frame, layered foam, fabric or leather upholsteryCare: Spot cleanSmall in stature but never short on personality, the Luva modular sectional has a squishy texture and a cartoonish shape. Available in a two and three seater configuration and countless upholstery options (like this Kindercore yellow), this versatile 78-inch sectional would feel right at home in a basement or a reading corner in need of some verve. It’s certainly a splurge, but for the iconic Herman Miller name, plus full assembly included in the price tag, it’s an investment that we imagine will hold up nicely over time.For a Modular ArrangementBurrow Range 3-Piece Sectional LoungerSpecsDimensions: 28"H x 62"W x 60.5"DSeat Depth: 21"Materials: Bent plywood frame, medium-density foam, shredded fill, fiber wadding, basketweave or velvet upholsteryCare: Water and bleach solutionInstead of opting for a giant L-shaped sectional that will take up the entire room, aim for something with a smaller footprint. Available in performance fabric and three leg finishes, this sectional couch is ideal for a tiny living space and a major upgrade from a smaller loveseat thanks to the chaise addition. While not an official fold-out sleeper sofa, the back cushions flop back to give you enough space should you be inclined to nap. Contributor David Kaufman, who tested it out for us, calls it “a great starter sofa for a first home or apartment that you won’t have to send to a landfill when you’re ready to move.”For Some TextureUrban Outfitters Macy Modular SofaSpecsDimensions: 30"H x 40"W x 40"DSeat Depth: Not listedMaterials: Polyester, plywood, pine woodCare: Spot cleanThis one is like a “choose your own adventure” in sofa form. Buy each piece separately to create the perfect sectional for your space. Corner chairs, seats with arms, seats without arms, an ottoman—you’ve got options. The ribbed corduroy is a throwback to college-dorm days (but in a good way).For Low-to-the-Ground LeatherArticle Cigar Rawhide Tan Reversible SectionalSpecsDimensions: 26"H x 92"W x 94"DSeat Depth: 27"Materials: Frame: kiln-dried solid pine, MDF, rubberwood; Filling: high-density foam, polyester fiber, duck feathers; Leather: 100% top grain, full-aniline Italian pull-up leatherCare: Wipe with a damp clothThe reversible chaise makes this sectional super versatile—meaning there’s a good chance it’ll fit right into your next space too. Low profile in silhouette and minimalist in style, the smooth leather texture carries this design’s personality while keeping things sturdy for movie nights. And while it may not be a sofa bed, don’t let anyone dissuade you from a full night’s snooze.Best sectionals for small spaces FAQs:What features should I look for in a sectional for small living rooms?Modular furniture—any sectional that comes in several parts so you can adjust your sofa to suit the configurations—is built with flexibility in mind, making it the ideal contender for a small living space. We have tested and loved Castlery modular sofas, but a search for modular sectional should give you plenty of other options. A room tight on space can benefit from couches with storage, like those from Albany Park. Now you’ll be able to minimize the clutter and extra blankets by stowing them underneath your seat.How do I arrange a sectional in a small space?Don’t fall for the tiniest sectional you can find, but rather invest in something well-made—it plays a big part in your living room, after all. While it may seem counterintuitive, Nashville interior designer Amanda Khouri says, “Filling a small room with small furniture can make it feel…well, smaller.”Designer Alicia Murphy used a sectional in the living room of a teensy trailer she converted into a plush family getaway in Montauk, New York, for this very reason: “If you use a corner sectional rather than a sofa and chair, you most likely will add seating for two or three more people,” Murphy explains, noting that the alternative prevents you from having about 36 inches of dead space in the corner.”

Banksy Unveils New Lighthouse Mural With the Words ‘I Want to Be What You Saw in Me’ in France The anonymous street artist announced the uncharacteristically personal artwork on May 29. It’s located on a wall in the French city of Marseille The new Banksy mural is on a wall beside a covered street in Marseille, France. Viken Kantarci / AFP via Getty Images After a five-month hiatus, Banksy is back with a new lighthouse mural in France. Compared to many of his previous works, the piece is uncharacteristically personal. The anonymous street artist announced his latest creation in an Instagram post on May 29. Located on a beige wall beside a sidewalk, the black lighthouse appears to merge with a painted shadow of a nearby metal safety post. Stenciled over the artwork are the words, “I want to be what you saw in me.” “There is no official explanation for the phrase,” write the Associated Press’ Thomas Adamson, Bishr Eltoni and Jill Lawless. “But its emotional pull is unmistakable—a quiet plea for recognition, love or redemption.”Banksy’s post appeared on Instagram with no caption, initially leaving the new mural’s location a mystery. However, BBC Verify, a team of investigative journalists and fact-checkers, quickly deduced that the mural is located on a backstreet called Rue Félix Frégier in the French city of Marseille, according to BBC News’ Cachella Smith and Amy Walker. Reactions to the new artwork were mixed. As the Art Newspaper’s Anny Shaw writes, “It could be a defection from the commercialism he simultaneously critiques and profits from, but I guess I’ve been made to expect more—either by technique or by message—from an artist whose entire practice, cheesy as it may be, is enmeshed in social and class consciousness.” Banksy has been creating street art around the world for decades. Some of his most well-known pieces include Rage, the Flower Thrower, a commentary on war, and Girl With Balloon. The artist’s works have sold for staggeringly high prices at auction; his most expensive piece is Love Is in the Bin, which fetched more than $25 million in 2021. Last summer, Banksy completed a series of animal-themed paintings in London, including a pair of elephants, monkeys swinging from a bridge and a howling wolf painted on a satellite dish. Until the lighthouse, his most recent verified mural was a black-and-white painting of a veiled woman breastfeeding a baby—a reference to the Madonna and child—which he announced in an Instagram post on December 16, 2024. Much of Banksy’s career has been “politically charged,” drawing attention to issues like Palestinian suffering and the refugee crisis, writes the Art Newspaper. For instance, his recent London animal artworks may have been “a comment on the impact of humans on biodiversity.” Swinger, a mural Banksy painted in New Orleans in 2008 Infrogmation of New Orleans via Wikimedia Commons under CC BY 2.0 However, the new lighthouse mural is a departure for Banksy, who “seems to be getting more philosophical in his middle age,” per the publication. His latest work could be “a moment of rare self-reflection—after all, this is the first time Banksy has referred to himself in the first person in a public mural.” Nobody knows what Banksy’s intentions were, though many onlookers have theories. The Guardian’s Matthew Weaver suggests that “I want to be what you saw in me” may be a reference to a lyric from “Softly,” a 2001 country song by the band Lonestar: “I want to be what you see in me / I want to love you the way that you love me.” Banksy rarely includes text in his pieces. When he does, he “usually incorporates extant signage or graffiti when adding his touch rather than writing out his own phrases,” according to Hyperallergic’s Rhea Nayyar. “That being said, this particular text takes the street artist’s already on-the-nose symbolism to another level,” with the “heavy-handed captioning” coming across as “overly sentimental.” “Some commentators have said that at best, the latest offering attempts to open a vaguely profound conversation about being a ‘guiding light’ or ‘finding the light within.’ At worst, it veers into ‘live, laugh, love’ territory,” writes the Art Newspaper. “This message could be Banksy’s most personal yet—but its execution risks it also being his most banal.” Get the latest stories in your inbox every weekday.

Greg Petronzi is a psychologist and a watchmaker. True Patina 2025-05-31T09:50:02Z Save Saved Read in app This story is available exclusively to Business Insider subscribers. Become an Insider and start reading now. Have an account? Greg Petronzi was a psychologist and professor when he got into watchmaking on the side. Now he repairs vintage Rolex watches and works on pieces worth six figures. Petronzi said his watchmaking out-earns his work in psychology. This as-told-to essay is based on a conversation with Greg Petronzi of True Patina, a watchmaker specializing in vintage Rolex repairs. He is also a licensed psychologist and professor at New York University. This interview has been edited for length and clarity.I've always had a fascination with watches, starting at a young age. Around middle school, I had a Swatch with a see-through plastic case and an automatic movement. You could see all of the parts of the watch as it was working, and I always found that so interesting.In college, I pursued psychology. That was my practical career path. I got a master's and a Ph.D. Overall, it was a 12-year path to becoming a licensed psychologist. But I never lost my interest in watches.After I finished school, I got into watchmaking, which ended up becoming my main gig. Psychology became my part-time gig, which I never anticipated happening, but it's super exciting. I love both disciplines.I got into watchmaking with an informal apprenticeshipWatchmaking was never really on my radar. As a grad student, I started getting involved in the world of watches and meeting other watch enthusiasts through forums and meetups.I made friends with a watchmaker out of Florida named Rik Dietel, who has 35-plus years of experience in watchmaking and specifically with vintage Rolex, which is the niche that I was very passionate about. I started asking him how to fix this or change that on my own watches. Little by little, Rik started teaching me, and it turned into an informal, remote apprenticeship.Over the next several years, while I was working as a psychologist and professor, my skillset started to really develop, and it started to become apparent to me that this might be more than just a hobby. I was working on my own stuff, then friends' watches, then friends of friends' watches. Then I started getting requests from people I didn't know. That's when I said, "I'd better take this a bit more seriously." Greg Petronzi in his workshop. True Patina During the pandemic, my psychology work went remote, and I was able to put more time and energy into watchmaking. I also had an income, which helped me afford the tools. I just recently spent about $15,000 on one tool to do a very nuanced repair.I built up my watchmaking workshop and started to develop a niche in cosmetic repair — dials and hands. Because I started as a watch collector, I understand the importance of preserving the originality of a watch. While a lot of more modern watchmakers have the disposition of "repair and replace," I have the disposition of "restore and retain."I started to showcase my work on Instagram and build trust in the community. I ended up working with some really important watch collectors and dealers, like Eric Wind of Wind Vintage, and auction houses like Phillips.There was a moment when I realized I'd made a name for myself in the watch world: Someone sent me an eBay listing that said, "Watch just serviced by True Patina." I had no idea who the seller was, but I thought, "Wow, my company name is actually carrying so much weight that people are using this as a flex toward selling their watch." That felt really good. Watchmaking can be challenging to get into, but it's really rewardingWatchmaking has more often than not outperformed what the Ph.D and psychology have allowed me to earn, which I'm astounded by and grateful for.If someone's interested in learning watchmaking, I'd say it's possible, but there are some challenges. The tools are expensive, and the formal education options are limited. Most people either find an apprenticeship or start by working for an established brand while slowly building up their own workshop.My pricing varies a lot based on how much restoration is needed. Repairs typically range between $1,000 and $2,000, but some go up to $6,000 or more, especially if a rare part needs to be sourced.I service watches that range in price from a few thousand dollars upward to six-figure watches. It's not uncommon for me to work on a watch that costs $200,000 and occasionally even up to $500,000. Most commonly, they are a few thousand up to $20,000. But what's most meaningful to me is working on sentimental pieces — watches that have been in families for generations.Becoming a watchmaker never crossed my mind as a kid or even as a college student. But all of a sudden, it organically became reality for me, and it's a very, very meaningful and enjoyable existence.Watchmaking is a very rewarding field. It can be very stressful, especially when the repair might not be cooperating the way you want it to, but when things do fall in line, it can be a very mindful activity. You get into this flow where time just sort of ceases. It's kind of ironic.Do you have a story to share about watches or watchmaking? Contact this reporter at kvlamis@businessinsider.com.