Autoregressive image generation has been shaped by advances in sequential modeling, originally seen in natural language processing. This field focuses on generating images one token at a time, similar to how sentences are constructed in language models. The appeal of this approach lies in its ability to maintain structural coherence across the image while allowing for high levels of control during the generation process. As researchers began to apply these techniques to visual data, they found that structured prediction not only preserved spatial integrity but also supported tasks like image manipulation and multimodal translation effectively. Despite these benefits, generating high-resolution images remains computationally expensive and slow. A primary issue is the number of tokens needed to represent complex visuals. Raster-scan methods that flatten 2D images into linear sequences require thousands of tokens for detailed images, resulting in long inference times and high memory consumption. Models like Infinity need over 10,000 tokens for a 1024×1024 image. This becomes unsustainable for real-time applications or when scaling to more extensive datasets. Reducing the token burden while preserving or improving output quality has become a pressing challenge. Efforts to mitigate token inflation have led to innovations like next-scale prediction seen in VAR and FlexVAR. These models create images by predicting progressively finer scales, which imitates the human tendency to sketch rough outlines before adding detail. However, they still rely on hundreds of tokens—680 in the case of VAR and FlexVAR for 256×256 images. Moreover, approaches like TiTok and FlexTok use 1D tokenization to compress spatial redundancy, but they often fail to scale efficiently. For example, FlexTok’s gFID increases from 1.9 at 32 tokens to 2.5 at 256 tokens, highlighting a degradation in output quality as the token count grows. Researchers from ByteDance introduced DetailFlow, a 1D autoregressive image generation framework. This method arranges token sequences from global to fine detail using a process called next-detail prediction. Unlike traditional 2D raster-scan or scale-based techniques, DetailFlow employs a 1D tokenizer trained on progressively degraded images. This design allows the model to prioritize foundational image structures before refining visual details. By mapping tokens directly to resolution levels, DetailFlow significantly reduces token requirements, enabling images to be generated in a semantically ordered, coarse-to-fine manner. The mechanism in DetailFlow centers on a 1D latent space where each token contributes incrementally more detail. Earlier tokens encode global features, while later tokens refine specific visual aspects. To train this, the researchers created a resolution mapping function that links token count to target resolution. During training, the model is exposed to images of varying quality levels and learns to predict progressively higher-resolution outputs as more tokens are introduced. It also implements parallel token prediction by grouping sequences and predicting entire sets at once. Since parallel prediction can introduce sampling errors, a self-correction mechanism was integrated. This system perturbs certain tokens during training and teaches subsequent tokens to compensate, ensuring that final images maintain structural and visual integrity. The results from the experiments on the ImageNet 256×256 benchmark were noteworthy. DetailFlow achieved a gFID score of 2.96 using only 128 tokens, outperforming VAR at 3.3 and FlexVAR at 3.05, both of which used 680 tokens. Even more impressive, DetailFlow-64 reached a gFID of 2.62 using 512 tokens. In terms of speed, it delivered nearly double the inference rate of VAR and FlexVAR. A further ablation study confirmed that the self-correction training and semantic ordering of tokens substantially improved output quality. For example, enabling self-correction dropped the gFID from 4.11 to 3.68 in one setting. These metrics demonstrate both higher quality and faster generation compared to established models. By focusing on semantic structure and reducing redundancy, DetailFlow presents a viable solution to long-standing issues in autoregressive image generation. The method’s coarse-to-fine approach, efficient parallel decoding, and ability to self-correct highlight how architectural innovations can address performance and scalability limitations. Through their structured use of 1D tokens, the researchers from ByteDance have demonstrated a model that maintains high image fidelity while significantly reducing computational load, making it a valuable addition to image synthesis research. Check out the Paper and GitHub Page. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 95k+ ML SubReddit and Subscribe to our Newsletter. NikhilNikhil is an intern consultant at Marktechpost. He is pursuing an integrated dual degree in Materials at the Indian Institute of Technology, Kharagpur. Nikhil is an AI/ML enthusiast who is always researching applications in fields like biomaterials and biomedical science. With a strong background in Material Science, he is exploring new advancements and creating opportunities to contribute.Nikhilhttps://www.marktechpost.com/author/nikhil0980/Teaching AI to Say ‘I Don’t Know’: A New Dataset Mitigates Hallucinations from Reinforcement FinetuningNikhilhttps://www.marktechpost.com/author/nikhil0980/This AI Paper Introduces LLaDA-V: A Purely Diffusion-Based Multimodal Large Language Model for Visual Instruction Tuning and Multimodal ReasoningNikhilhttps://www.marktechpost.com/author/nikhil0980/NVIDIA AI Introduces Fast-dLLM: A Training-Free Framework That Brings KV Caching and Parallel Decoding to Diffusion LLMsNikhilhttps://www.marktechpost.com/author/nikhil0980/Meet NovelSeek: A Unified Multi-Agent Framework for Autonomous Scientific Research from Hypothesis Generation to Experimental Validation

Architects: Want to have your project featured? Showcase your work by uploading projects to Architizer and sign up for our inspirational newsletters.   Is an architecture firm designing its own studio the equivalent of an artist painting a self-portrait? (Should we coin the term “auto-architecture?”) Perhaps this isn’t a perfect analogy, but it certainly contains parallels that are productive to parse… Studio spaces are distinct from offices in that they not only shape daily rituals and structure relationships between colleagues but also act as an expression of the values at the core of the firm’s design philosophies. Freed from the usual constraints of client briefs, for many firms, designing their own workspace offers a unique opportunity for experimentation and self-expression. The studios featured in this collection span diverse geographies and contexts — from a vaulted school library repurposed as an “anti-office,” to a carbon-neutral warehouse conversion in Sydney, to a minimalist tiled atelier in Casablanca. Despite their differences, each workspace shares a commitment to thoughtful design that blurs the line between functions and offers a vision for cultivating creativity. More than places of production, these studios are active expressions of architectural identity; spaces that support not only what architects make, but how they make it. They also challenge outdated typologies and embrace the hybrid realities of contemporary practice. Skylab HQ By Skylab, Portland, Oregon After spending years in a historic structure in downtown Portland, the Skylab team decided the time had come to create a space that reflected the dynamic nature of their practice. They asked themselves: “How can our studio evolve from a dedicated workspace to a playground for the art and design community? Where can we find a space to integrate gardens, an event venue, and a fabrication shop, as well as our studio?” Leaving the downtown core, they opted to transform a pair of WWII-era prefabricated steel warehouses into a hybrid studio, fabrication lab and cultural venue supporting both architectural production and artistic exchange. Strategic insertions — like a 60-foot-long (18-meter) ridge skylight, 10-foot (3-meter) operable window walls and CLT-framed meeting rooms — maximize daylight and material contrast, balancing industrial grit with biophilic warmth. The adaptive reuse reflects the firm’s ethos of experimentation, extending their design process into the very architecture that houses it. Alexander House By Alexander &CO., Sydney, Australia Jury Winner, Architecture +Workspace, 10th Annual A+Awards Alexander House functions as both studio and experimental prototype, integrating low-carbon construction with hybrid live/work spatial typologies tailored to an evolving architectural practice. While functioning as an architectural residential showcase, the team also works from this home, and their clients meet with them there; the project challenges preconceptions of home, land, family and work. From a voluminous material library in the basement to a concrete mezzanine bench designed for quiet focus, the layout supports varied modes of design work while challenging conventional boundaries between domestic and professional space. Crafted in collaboration with local makers, the building also pioneers sustainability through reclaimed timber linings, carbon-neutral bricks, and a solar system supplying up to 80% of daily energy demand. say architects Community Office By say architects, Hangzhou, China Say Architects’ office reimagines workplace architecture as a life-oriented, materially expressive environment, where exposed I-beams structure both the building and the studio’s daily rhythms. Cantilevered volumes, rope-grown greenery, and integrated misting systems animate the exterior, while steel-framed shelving and model rooms of rich timber textures create a tactile, inspiration-driven interior. Prioritizing adaptability and sensory comfort, the space dissolves traditional partitions in favor of spatial arrangements that align with design habits, offering a studio that is both tool and manifesto. Bohlin Cywinski Jackson, Philadelphia Studio By Bohlin Cywinski Jackson, Philadelphia, Pennsylvania Bohlin Cywinski Jackson’s Philadelphia studio transforms a historic social club (founded in 1923) into a contemporary workspace through adaptive reuse, prioritizing flexibility, daylight and material economy. The goal was to create a highly flexible work environment that would allow designers to move quickly between individual work, impromptu discussions and group meetings throughout the day. Restored terrazzo floors and ornamental detailing anchor a modern layout featuring hoteling desks, collaborative mezzanine zones and panoramic views of the city center.  The design supports agile workflows and hybrid collaboration while integrating repurposed custom furnishings to extend the life cycle of past projects. ADND OFFICE By Atelier Design N Domain, Mumbai, India ADND’s new Bombay headquarters is a richly layered adaptive reuse of a century-old industrial warehouse, reimagined as an expressive design laboratory charged with material experimentation and symbolic nuance. The studio’s soaring central bay reaches 26 feet (8 meters) in height, punctuated by 7-foot (2-meter) pivoting porthole windows that flood the workspace with southern light, evoking a cathedral-like ambiance. Throughout, bespoke interventions — from terrazzo-cast floors and mirrored reception desks to hand-sketched upholstery and looped oak chairs — translate the founders’ personal design dialects into architectural form, creating a space where industrial memory and contemporary authorship converge. Studio Cays X Studio BO By Studio CAYS, Casablanca, Morocco In this Casablanca-based studio, minimalist rigor meets material clarity through tiled walls and seamless epoxy flooring, crafting a luminous, low-maintenance workspace. At its core, a central bench anchors the open-plan layout, fostering daily collaboration and reinforcing the studio’s emphasis on shared ideation within a purified architectural envelope. Smart Design Studio By smart design studio, Alexandria, Australia Jury Winner, Office Interiors (<25,000 sq ft); Jury Winner, Office Building Low Rise, 10th Annual A+Awards Smart Design Studio’s headquarters fuses industrial heritage with cutting-edge sustainability, transforming a conserved warehouse into a carbon-neutral workspace powered by on-site energy and water collection systems. The studio’s open-plan interior is crowned by a mezzanine framed by original steel trusses, while a striking vaulted residence above features self-supporting brick catenary arches — an elegant synthesis of structural economy and sculptural ambition. Designed to reflect the material restraint and innovation of early industrial architecture, the building is a working manifesto for the studio’s interdisciplinary ethos. Architect’s Office at Kim Yam Road By Park + Associates, Singapore Popular Choice Winner, Office Interiors, 10th Annual A+Awards Photos by Edward Hendricks Occupying a former library hall atop a repurposed 1960s school, this studio embraces the latent grandeur of its barrel-vaulted, column-free volume to craft a boundary-less, anti-office environment. Full-height louvered windows invite daylight and breeze through the arching space, while the design resists conventional programming in favor of layered, informal settings that foster creativity and fluid collaboration. Rather than overwrite its past, the intervention amplifies the building’s inherent spatial expression; through adaptive reuse, the architects position atmosphere as architecture. Architects: Want to have your project featured? Showcase your work by uploading projects to Architizer and sign up for our inspirational newsletters.   The post Self-Portrait in Plan: 8 Architecture Studios Designed By Their Owners appeared first on Journal.

“Was your interior designer Ayn Bland?” Barely ten minutes into Mountainhead, the first feature film from Succession creator Jesse Armstrong, the viewer is confronted with one of the film’s central tenets: beige is bad.Articulated by Jeff (Ramy Youssef), the movie does a lot to engage this train of thought. Its central characters—tech bro founders Randall (Steve Carrell), Hugo (Jason Schwartzman), and Venis (Cory Michael Smith)—are pitted against not just each other, but the world writ large as it begins to crumble around them, mostly thanks to their own machinations.But when did our current beige malaise set in? Much of the recent divisiveness might be attributed to the Kardashians, whose homes are synonymous with muted putty tones: “less is more” taken to its only logical conclusion—least is best. The aesthetic has occasionally gone viral, with the TikTok account Sad Beige even garnering notoriety for documenting the lifelessness of children’s retail offerings in the voice of German documentary filmmaker Werner Herzog.Warner Bros.The cast of Mountainhead, the new HBO Original film from Succession creator Jesse Armstrong.Social media is as much a progenitor of the trend as it is the platform to document it and repackage it for frictionless consumption. Beige is safe, easy, legible—if your entire house has been unpacked from Amazon boxes, why fight the natural inclination to just live in one? The Mountainhead home might be the lair of a fictional tech boss, but the aesthetics remain the same up and down the corporate ladder.The color family has its own troubled history: “khaki,” which is derived from the Urdu term for “dusty,” originated in mid-19th-century colonial India, as the critic and author Andrea Codrington noted in an eerily prescient 2001 essay for Cabinet magazine that further elaborates on beige’s latent potential for evil. Mountainhead, then, sees the hue returning to its roots, becoming shorthand for what Armstrong sees as everything that’s wrong with the world: complacency, exhaustion, and boredom.MACALL POLAY. SMPSPAnother beige couch in the Mountainhead house. Armstrong’s camera treats the home, designed by local Utah firm Upwall Design, as a character in and of itself, with lingering shots of quietly humming servers, spiraling staircases and vacant driveways presenting solitude and menace as two sides of the same coin. Town & Country reported that production designer Stephen Carter liked the property for its The Shining-esque vibes. He wasn’t wrong. “There’s a solitary nature to that house, too,” star Steve Carrell told The Salt Lake Tribune. “You feel like you’re away from everything.”Another word for that feeling—articulated by scenes of the men laying about the home glued to their phones as scenes of worldwide terror stream in—would be dissociation. Think home is where the heart is? Mountainhead makes a compelling case that it's often just another a bad trip.Sean SantiagoDeputy EditorSean Santiago is ELLE Decor's Deputy Editor, covering news, trends and talents in interior design, hospitality and travel, culture, and luxury shopping. Since starting his career at an interior design firm in 2011, he has gone on to cover the industry for Vogue, Architectural Digest, Sight Unseen, PIN-UP and Domino. He is the author of The Lonny Home (Weldon Owens, 2018), has produced scripted social content for brands including West Elm and Streeteasy, and is sometimes recognized on the street for his Instagram Reels series, #DanceToDecor

Basic scientific research is a key contributor to economic productivity.getty Is science running out of steam? A growing body of research suggests that disruptive breakthroughs—the kind that fundamentally redefine entire fields—may be occurring less frequently. A 2023 article in Nature reported that scientific papers and patents are, on average, less “disruptive” than they were in the mid-20th century. The study sparked intense interest and considerable controversy, covered in a recent news feature provocatively titled “Are Groundbreaking Science Discoveries Becoming Harder To Find?” Before weighing in, however, it is worth interrogating a more fundamental question: What do we mean when we call science “disruptive”? And is that, in fact, the appropriate benchmark for progress? The study in question, led by entrepreneurship scholar Russell Funk, employs a citation-based metric known as the Consolidation–Disruption (CD) index. The tool attempts to quantify whether new research displaces prior work—a signal of disruption—or builds directly upon it, thereby reinforcing existing paradigms. It represents a noteworthy contribution to our understanding of scientific change. Their conclusion, that disruption has declined across disciplines even as the volume of scientific output has expanded, has ignited debate among scientists, scholars and policymakers. Innovation May Be Getting Harder—But Also Deeper At a structural level, science becomes more complex as it matures. In some sense it has to slow down. The simplest questions are often the first to be answered, and what remains are challenges that are more subtle, more interdependent, and more difficult to resolve. The law of diminishing marginal returns, long familiar in economics, finds a natural corollary in research: at some point the intellectual “low-hanging fruit” has largely been harvested. Yet this does not necessarily imply stagnation. In fact, science itself is evolving. I think that apparent declines in disruption reflect not an impoverishment of ideas, but a transformation in the conduct and culture of research itself. Citation practices have shifted. Publication incentives have changed. The sheer availability of data and digital resources has exploded. Comparing contemporary citation behavior to that of earlier decades is not simply apples to oranges; it’s more like comparing ecosystems separated by tectonic time. More profoundly, we might ask whether paradigm shifts—particularly those in the Kuhnian sense—are truly the milestones we should prize above all others. Much of the innovation that drives societal progress and economic productivity does not emerge from revolutions in thought, but from the subtle extension and application of existing knowledge. In fields as varied as biomedicine, agriculture, and climate science, incremental refinement has yielded results of transformative impact.Brighter green hybrid rice plants (left) help increase yields at this Filipino farm. (Photo by ... More Dick Swanson/Getty Images)Getty Images Science Today Is More Sophisticated—And More Efficient Scientists are publishing more today than ever. Critics of contemporary science attribute this to metric-driven culture of “salami slicing,” in which ideas are fragmented into the “minimum publishable unit” so that scientists can accrue an ever-growing publication count to secure career viability in a publish-or-perish environment. But such critiques overlook the extraordinary gains in research efficiency that have occurred in the past few decades, which I think are a far more compelling explanation for the massive output of scientific research today. Since the 1980s, personal computing has transformed nearly every dimension of the scientific process. Manuscript preparation, once the province of typewriters and retyped drafts, has become seamless. Data acquisition now involves automated sensors and real-time monitoring. Analytical tools like Python and R allow researchers to conduct sophisticated modeling and statistics with unprecedented speed. Communication is instantaneous. Knowledge-sharing platforms and open-access journals have dismantled many of the old barriers to entry.Advances in microcomputer technology in the 1980s and 1990s dramatically accelerated scientific ... More research.Denver Post via Getty Images Indeed, one wonders whether critics have recently read a research paper from the 1930s or 1970s. The methodological rigor, analytical depth, and interdisciplinary scope of modern research are, by nearly any standard, vastly more advanced. The Horizon Has Expanded In biology alone, high-throughput technologies—part of the broader “omics” revolution catalyzed by innovations like the polymerase chain reaction (PCR), which enabled rapid DNA amplification and supported the eventual success of the Human Genome Project—continue to propel discovery at an astonishing pace.Nobel Prize laureate James D. Watson speaks at a press conference to announce that a six-country ... More consortium has successfully drawn up a complete map of the human genome, completing one of the most ambitious scientific projects ever and offering a major opportunity for medical advances, 14 April 2003 at the National Institute of Health in Bethesda, Maryland. The announcement coincides with the 50th anniversary of the publication of the landmark paper describing DNA's double helix by Watson and Francis Crick. AFP PHOTO / Robyn BECK (Photo credit should read ROBYN BECK/AFP via Getty Images)AFP via Getty Images When critics lament the apparent decline of Nobel-caliber “blockbusters” they overlook that the frontier of science has expanded—not narrowed. If we consider scientific knowledge as a volume, then it is bounded by an outer edge where discovery occurs. In Euclidean geometry, as the radius of a sphere increases, the surface area (scaling with the square of the radius) grows more slowly than the volume (which scales with the cube). While the volume of knowledge grows more rapidly—encompassing established theories and tools that continue to yield applications—the surface area also expands, and it is along this widening frontier, where the known meets the unknown, that innovation arises. Rethinking Returns on Investment The modern belief that science must deliver measurable economic returns is, historically speaking, a relatively recent development. Before the Second World War, scientific research was not broadly viewed as a driver of productivity. Economist Daniel Susskind has argued that even the concept of economic growth as a central policy goal is a mid-20th century invention. After the war, that changed dramatically. Governments began to see research as critical to national development, security, and public health. Yet even as expectations have grown, relative public investment in science has, paradoxically, diminished, despite the fact that basic scientific research is a massive accelerant of economic productivity and effectively self-financing. While absolute funding has increased, government spending on science as a share of GDP has declined in the US and many other countries. Given the scale and complexity of the challenges we now face, we may be underinvesting in the very enterprise that could deliver solutions. Recent proposals to cut funding for NIH and NSF could, by some estimates, cost the U.S. tens of billions in lost productivity. There is compelling evidence to suggest that significantly increasing R&D expenditures—doubling or even tripling them—would yield strong and sustained returns. AI and the Next Wave of Scientific Efficiency Looking to the future, artificial intelligence offers the potential to not only streamline research but also to augment the process of innovation itself. AI tools—from large language models like ChatGPT to specialized engines for data mining and synthesis—enable researchers to traverse disciplines, identify patterns, and generate new hypotheses with remarkable speed. The ability to navigate vast bodies of scientific literature—once reserved for those with access to elite research libraries and ample time for reading—has been radically democratized. Scientists today can access digitized repositories, annotate papers with precision tools, manage bibliographies with software, and instantly trace the intellectual lineage of ideas. AI-powered tools support researchers in sifting through and synthesizing material across disciplines, helping to identify patterns, highlight connections, and bring under-explored ideas into view. For researchers like myself—an ecologist who often draws inspiration from nonlinear dynamics, statistical physics, and cognitive psychology—these technologies function as accelerators of thought rather than substitutes for it. They support the process of discovering latent analogies and assembling novel constellations of insight, the kind of cognitive recombination that underlies true creativity. While deep understanding still demands sustained intellectual engagement—reading, interpretation, and critical analysis—these tools lower the barrier to discovery and expand the range of intellectual possibilities. By enhancing cross-disciplinary thinking and reducing the latency between idea and investigation, AI may well reignite the kind of scientific innovation that some believe is slipping from reach. Science as a Cultural Endeavor Finally, it bears emphasizing that the value of science is not solely, or even primarily, economic. Like the arts, literature, or philosophy, science is a cultural and intellectual enterprise. It is an expression of curiosity, a vehicle for collective self-understanding, and a means of situating ourselves within the universe. From my vantage point, and that of many colleagues, the current landscape of discovery feels more fertile than ever. The questions we pose are more ambitious, the tools at our disposal more refined, and the connections we are able to make more multidimensional. If the signal of disruption appears to be dimming, perhaps it is only because the spectrum of science has grown too broad for any single wavelength to dominate. Rather than lament an apparent slowdown, we might ask a more constructive question: Are we measuring the right things? And are we creating the conditions that allow the most vital forms of science—creative, integrative, and with the potential to transform human society for the better—to flourish?

Office P&L / Studio TngtetshiuSave this picture!© Studio MillspaceOffices, Renovation•Tainan, Taiwan Architects: Studio Tngtetshiu Year Completion year of this architecture project Year:  2024 Photographs Photographs:Studio MillspaceMore SpecsLess Specs Save this picture! Text description provided by the architects. This project begins with a priority on economy—working with, rather than against, the existing spatial typologies and structural conditions. By minimizing design intervention, the intent was to underscore and reveal the latent character of the original architecture while introducing new office functions. As the new program overlays the old structural order, the original spatial qualities are not erased but reawakened.Save this picture!Save this picture!Save this picture!Save this picture!Save this picture!The site comprises two warehouse buildings flanking a transitional space. Each structure employs distinct systems—timber trusses, reinforced concrete, and steel framing—coexisting in close proximity. This juxtaposition of temporal and spatial languages creates a compelling heterogeneity. This layered composition quietly echoes the city's complex and multifaceted history.Save this picture!Save this picture!Save this picture!The in-between space, with its semi-exterior qualities, connects the two primary volumes and offers a layered experience where interior and exterior perceptions intertwine. A raised floor unifies the interior levels, while portions of the original concrete surface —once exterior— are deliberately preserved, forming an indoor courtyard that retains an outdoor sensibility. This courtyard acts as an intermediary between the workspace and the pantry, blurring the thresholds of function and atmosphere.Save this picture!Save this picture!Approaching from the entrance steps, the eye is drawn by an 18-meter-long serpentine pendant light, which leads through a narrow, irregular corridor. This meandering path resists a single glance, unfolding gradually through the courtyard and into the main working area. Here, work tables and display shelving run along the longitudinal axis, emphasizing spatial depth. Adjustable curtains define zones and modulate openness, guiding the eye naturally toward the meeting room at the far end—a light-filled space framed by windows on three sides.Save this picture!Rather than pursuing a singular, unified aesthetic, the design embraces strategies of bricolage. Fragmentary architectural elements and ambiguous objects—hovering between old and new—are arranged in overlapping layers. Through movement and occupation, users are invited to participate in an ongoing narrative of time, memory, and place.Save this picture! Project gallerySee allShow less Project locationAddress:Tainan, TaiwanLocation to be used only as a reference. It could indicate city/country but not exact address.About this officeStudio TngtetshiuOffice••• Published on May 23, 2025Cite: "Office P&L / Studio Tngtetshiu" 23 May 2025. ArchDaily. Accessed . <https://www.archdaily.com/1030346/office-p-and-l-studio-tngtetshiu&gt ISSN 0719-8884Save世界上最受欢迎的建筑网站现已推出你的母语版本!想浏览ArchDaily中国吗?是否 You've started following your first account!Did you know?You'll now receive updates based on what you follow! Personalize your stream and start following your favorite authors, offices and users.Go to my stream

Large Reasoning Models (LRMs) like OpenAI’s o1 and o3, DeepSeek-R1, Grok 3.5, and Gemini 2.5 Pro have shown strong capabilities in long CoT reasoning, often displaying advanced behaviors such as self-correction, backtracking, and verification—collectively known as “aha moments.” These behaviors have been observed to emerge through outcome-driven RL without the need for supervised fine-tuning. Models like DeepSeek-R1 and its open-source replications (e.g., TinyZero and Logic-RL) have demonstrated that carefully designed RL pipelines—using rule-based rewards, curriculum learning, and structured training—can induce such reflective reasoning abilities. However, these emergent behaviors tend to be unpredictable and inconsistent, limiting their practical reliability and scalability. To address this, researchers have explored structured RL frameworks that target specific reasoning types, such as deduction, abduction, and induction. These approaches involve aligning specialist models, merging them in parameter space, and applying domain-specific continual RL. Tools like Logic-RL use rule-conditioned RL to solve logic puzzles, improving transferability to tasks like math reasoning. Meanwhile, other works propose mechanisms to enhance reasoning robustness, such as training models to reason both forwards and backwards, or iteratively self-critiquing their outputs. Studies analyzing “aha moments” suggest that these behaviors stem from internal shifts in uncertainty, latent representation, and self-assessment, offering new insights into engineering more reliable reasoning models.  Researchers from the National University of Singapore, Tsinghua University, and Salesforce AI Research address the limitations of relying on spontaneous “aha moments” in large language models by explicitly aligning them with three core reasoning abilities: deduction, induction, and abduction. They introduce a three-stage pipeline—individual meta-ability alignment, parameter-space merging, and domain-specific reinforcement learning—significantly enhancing model performance. Using a programmatically generated, self-verifiable task suite, their approach boosts accuracy over instruction-tuned baselines by over 10%, with further gains from domain-specific RL. This structured alignment framework offers a scalable, generalizable method for improving reasoning across math, coding, and science domains.  The researchers designed tasks aligned with deduction, induction, and abduction by using a structured “given two, infer the third” format based on hypothesis (H), rule (R), and observation (O). Deduction is framed as satisfiability checking, induction as masked-sequence prediction, and abduction as reverse rule-graph inference. These tasks are synthetically generated and automatically verified. The training pipeline includes three stages: (A) independently training models for each reasoning type using REINFORCE++ with structured rewards, (B) merging models through weighted parameter interpolation, and (C) fine-tuning the unified model on domain-specific data via reinforcement learning, isolating the benefit of meta-ability alignment.  The study evaluates models aligned with meta-abilities—deduction, induction, and abduction—using a curriculum learning setup across difficulty levels. Models trained on synthetic tasks strongly generalize to seven unseen math, code, and science benchmarks. At both 7B and 32B scales, meta-ability–aligned and merged models consistently outperform instruction-tuned baselines, with the merged model offering the highest gains. Continued domain-specific RL from these merged checkpoints (Domain-RL-Meta) leads to further improvements over standard RL finetuning (Domain-RL-Ins), especially in math benchmarks. Overall, the alignment strategy enhances reasoning abilities, and its benefits scale with model size, significantly boosting performance ceilings across tasks.  In conclusion, the study shows that large reasoning models can develop advanced problem-solving skills without depending on unpredictable “aha moments.” By aligning models with three core reasoning abilities—deduction, induction, and abduction—using self-verifiable tasks, the authors create specialist agents that can be effectively combined into a single model. This merged model outperforms instruction-tuned baselines by over 10% on diagnostic tasks and up to 2% on real-world benchmarks. When used as a starting point for domain-specific reinforcement learning, it raises performance by another 4%. This modular, systematic training approach offers a scalable and controllable foundation for building reliable, interpretable reasoning systems.  Check out the Paper and GitHub Page. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 95k+ ML SubReddit and Subscribe to our Newsletter. Sana HassanSana Hassan, a consulting intern at Marktechpost and dual-degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. With a keen interest in solving practical problems, he brings a fresh perspective to the intersection of AI and real-life solutions.Sana Hassanhttps://www.marktechpost.com/author/sana-hassan/RXTX: A Machine Learning-Guided Algorithm for Efficient Structured Matrix MultiplicationSana Hassanhttps://www.marktechpost.com/author/sana-hassan/From Protocol to Production: How Model Context Protocol (MCP) Gateways Enable Secure, Scalable, and Seamless AI Integrations Across EnterprisesSana Hassanhttps://www.marktechpost.com/author/sana-hassan/Researchers from Renmin University and Huawei Propose MemEngine: A Unified Modular AI Library for Customizing Memory in LLM-Based AgentsSana Hassanhttps://www.marktechpost.com/author/sana-hassan/Meta Introduces KernelLLM: An 8B LLM that Translates PyTorch Modules into Efficient Triton GPU Kernels

Villino Liberty Trastevere / STUDIOTAMATSave this picture!© Serena Eller - Ellerstudio Architects: STUDIOTAMAT Area Area of this architecture project Area:  110 m² Year Completion year of this architecture project Year:  2024 Photographs Photographs:Serena Eller - Ellerstudio Manufacturers Brands with products used in this architecture project Manufacturers:  Campeggi, Artceram, Artemide, Axolight, Cielo, DEMA, Danese Milano, Estiluz, Ex.T, Foster, Galassia, Irsap, La Pietra Compattata, Mutina, Oscar ono, Popham Design, Quadro Design, Tubes Radiatori, VippMore SpecsLess Specs Save this picture! Text description provided by the architects. In the heart of Rome's Trastevere district, STUDIOTAMAT has breathed new life into a hidden Liberty-style gem tucked away in the courtyard of a late 19th-century building along Viale di Trastevere. Once the caretaker's house for the old train station—or a neighborhood doctor's office, depending on who you ask—this forgotten villino has been reimagined as a refined urban hideaway for two. Spread across three compact, layered floors occupying 80 square meters and culminating in a lush, green terrace, the project began by carefully preserving the home's most distinctive elements. Restoring the front veranda with its delicate cathedral glass in soft greens, pinks, and yellows called for a mix of craft and technical precision. The original rhythm and hues were respected, while the frame was rebuilt in steel and solar-control glass. By removing the old French door that once divided it from the house, the veranda now flows into the interior, extending the living space and bathing it in natural light that subtly shifts in tone throughout the day.Save this picture!Save this picture!Save this picture!Save this picture!Inside, the intervention focuses on reconfiguring the layout, previously fragmented by a tight spiral staircase, and on the perception of the spaces. The redesign is radical in gesture but sensitive in execution: by moving the kitchenette beside the veranda, space is opened up for a striking alternating-tread staircase in chestnut wood. Its first step, clad in Verde Alpi marble, becomes the sculptural base of a custom bookshelf built into the understairs. Nearby, a mirrored chestnut storage unit conceals the laundry and enhances the sense of openness. The living room gains new depth, framed by a soaring double-height window that looks out onto surrounding gardens, in quiet harmony with Munari's iconic Falkland pendant lamps. The custom kitchen, liberated from overhead cabinetry, is defined by a linear base topped in Verde Alpi marble and shaded drawers that fade from black to terracotta, echoing the beautifully preserved original terracotta floors. A deep blue volume organizes the ground floor's services: fridge and pantry on one side, a discreet powder room on the other.Save this picture!This bold block of color continues upward, passing through the mezzanine and defining the main bathroom on the top floor, where Nouveau furnishings by Ex.t meet the textured surfaces of Patricia Urquiola's Mater tiles for Mutina and fixtures by Formafantasma for Quadro Design. To bring in light and create a sense of airiness, the mezzanine has been partially opened, introducing a double-height void. Glass floor panels offer glimpses between levels, while mirrored panels below bounce reflections upward, visually expanding the space and enhancing the blue volume. The remaining floors are finished in Foret parquet by Oscar Ono Paris, designed by Raphael Navot—preassembled oak slats with visible end grain, a nod to the pebble streets of 19th-century Paris and ancient Rome. A custom bed with drawer base and integrated headboard echoes the kitchen's color gradient, creating visual continuity between the levels.Save this picture!Save this picture!Save this picture!A second spiral staircase in raw iron with cherry wood treads connects the sleeping area to the upper level, serving as a sculptural focal point for the studio space. A glass partition, echoing the veranda's rhythm with alternating clear and ribbed panels, elegantly screens the bathroom: the door doubles as a backdrop for the shower, while a Verde Alpi marble sink slices through the glass, becoming a shared counter surface. Outside, the terrace features compacted stone paving in two shades of green, blending into the surrounding vegetation and reflecting the palette.Save this picture!Save this picture!Discreetly nestled in one of Rome's most characterful neighborhoods, this intervention captures the quiet charm of Trastevere. Just like the district itself, where modest façades often hide unexpected treasures, this project reveals the latent beauty of a neglected house. Merging memory with material, and history with contemporary sensibility, STUDIOTAMAT has created a deeply rooted, yet strikingly modern urban refuge.Save this picture! Project gallerySee allShow less About this officeSTUDIOTAMATOffice••• MaterialGlassMaterials and TagsPublished on May 22, 2025Cite: "Villino Liberty Trastevere / STUDIOTAMAT" 22 May 2025. ArchDaily. Accessed . <https://www.archdaily.com/1030344/villino-liberty-trastevere-studiotamat&gt ISSN 0719-8884Save世界上最受欢迎的建筑网站现已推出你的母语版本!想浏览ArchDaily中国吗?是否 You've started following your first account!Did you know?You'll now receive updates based on what you follow! Personalize your stream and start following your favorite authors, offices and users.Go to my stream