• Southern Californias water crisis is fueling fires. These maps show just how dry the region is
    www.fastcompany.com
    Dry conditions across Southern California set the stage for a series of deadly wind-driven wildfires that burned thousands of homes and other structures in the Los Angeles area in early January 2025.Ming Pan, a hydrologist at the University of California-San Diegos Center for Western Weather and Water Extremes, tracks the states water supplies. He put Southern Californias dryness into perspective using charts and maps.How dry is Southern California right now?In early January, the soil moisture in much of Southern California was in the bottom 2% of historical records for that day in the region. Thats extremely low.Hydrologists in California watch the sky very closely starting in October, when Californias water year begins.On January 8, 2025, the soil moisture content, as measured down to about 40 inches, was in the bottom 2% historically for that day in the area around Los Angeles. [Image: NASA]The state gets very little rain from May through September, so late fall and winter are crucial to fill reservoirs and to build up the snowpack to provide water. California relies on the Sierra snowpack for about one-third of its freshwater supply.However, Southern California started out the 2024-25 water year pretty dry. The region got some rain from an atmospheric river in November, but not much. After that, most of the atmospheric rivers that hit the West Coast from October into January veered northward into Washington, Oregon, and Northern California instead.When the air is warm and dry, transpiration and evaporation also suck water out of the plants and soil. That leaves dry vegetation that can provide fuel for flying embers to spread wildfires, as the Los Angeles area saw in early January.Water year data from October 1, 2024, to January 7, 2025, shows precipitation levels and the anomaly from the 1991-2020 average. [Image: Center for Western Weather and Water Extremes]So, while Northern Californias water and snowpack conditions are in good shape, Southern California is much drier and its water storage is not doing so well.The Southern Sierra snowpack was starting to dip below normal in early January.How snowpack and reservoir levels as of January 7, 2025, compare with the 2000-2015 average in Northern and Southern California: The shaded areas show normal reservoir levels (blue) and reservoir levels plus snowpack (gray). The lines track each for the 2025 water year. [Image: Center for Western Weather and Water Extremes]What can California expect for the rest of 2025?The U.S. Climate Prediction Centers seasonal outlook through March suggests that drought is likely to develop in the region in the coming months.The outlook takes into account forecasts for La Nia, an ocean temperature pattern that was on its way to developing in the Pacific Ocean in early 2025. La Nia tends to mean drier conditions in Southern California. However, not every La Nia affects California in the same way.One or two big rain events could completely turn the table for Southern Californias water situation. In 2023, California saw atmospheric rivers in April.So, its hard to say this early in the season how dry Southern California will be in the coming months, but its clear that people in dry areas need to pay attention to the risks.Ming Pan is a senior research hydrologist at the University of California, San Diego.This article is republished from The Conversation under a Creative Commons license. Read the original article.
    0 Σχόλια ·0 Μοιράστηκε ·142 Views
  • Ryue Nishizawa's Moriyama House was the most significant building of 2005
    www.dezeen.com
    Next up in our 21st-Century Architecture: 25 Years 25 Buildings series, we take a look at the most significant building of 2005, Moriyama House in Japan by architect Ryue Nishizawa.For those unfamiliar, Moriyama House can be quite quickly summarised by artist Henk Visch's experience of it in 2007, when he visited to create sculptures for its owner, Yasuo Moriyama."When visiting Mr Moriyama's house to find a spot for my sculptures, I could hardly describe where I was," Visch recalled."This was no normal house. Where was the front door?" he asked.Ryue Nishizawa's Moriyama House was the most significant building of 2005Nestled on a compact plot in the dense neighbourhood of Kamata in Tokyo, Moriyama House is a network of minimalist, non-hierarchical blocks stitched together with tiny gardens.Its design by Japanese architect Ryue Nishizawa the co-founder of Pritzker Architecture Prize-winner SANAA rewrote all the typical rules of privacy, thresholds and density when it comes to housing.Though describing the project as a house, or even a building, feels like a disservice to its design. Labelling it as a miniature city or a microcosm of Tokyo is somewhat more accurate.Moriyama House is made up of ten white blocksMoriyama House is made up of 10 white blocks, ranging in height from one to three storeys and occupying just half of its 290-square-metre plot.The blocks each serve a unique function, requiring residents to step outside as they move between the spaces, just as though they are navigating a mini village.Between them are paths and gardens that link directly to the surrounding streets, free of fencing and leaving the boundary between private and public space ambiguous.Nishizawa's tactic is reminiscent of the layering of ancient Japanese buildings Rob Gregory in The Architectural ReviewThis feeling of openness to the public helps to establish an unusual sense of community at the site, which is enhanced by the placement of large square windows allowing views into the white boxes from all around an unusual feature in a Japanese house, where privacy is usually prioritised."Japanese people often like a fence to enclose the property," Nishizawa once said when discussingthe design."But this project, there's no fence to define the property. Anybody in this area can get in and out."Minimalist white finishes unify the individual buildingsThe fragmented layout mimics the surrounding urban fabric, in which individual homes fill their plot but do not touch the neighbouring structures, leaving small walkways between them.It also reflects Nishizawa's view that "life can't be contained within a single lot"."People's sense of living expands beyond it, effectively erasing all borders," he once said.Read: Herzog & de Meuron's Tate Modern was the most significant building of 2000After visiting the house in 2007, former senior editor at The Architectural Review Rob Gregory said Nishizawa's design is "reminiscent of the layering of ancient Japanese buildings"."This striking group of six dwellings is a demonstration of how traditional notions of privacy and community can be adjusted," Gregory wrote."The Moriyama House has echoes of primitive forms of dwelling, where functions were distributed as isolated units that collectively create a defensible settlement. It also addresses Nishizawa's key concern regarding how to open up the house as a component of the city," he continued."Critical of the fact that many Japanese houses have become too hermetic and opaque, with homes that increasingly turn away from the street to focus on internal courts, Nishizawa's tactic is reminiscent of the layering of ancient Japanese buildings."The blocks each serve a unique functionCurrently, the owner occupies one of the volumes, which contains two bedrooms, a study and a living area. The rest of them are used as rental apartments.Despite their different functions, the blocks are unified by their bright and geometric designs, formed of thin steel sheet panels that lend a bold minimalist look to the site.In tandem with access to greenery and wide open windows that maximise natural light, this deliberate simplicity helps create a sense of expansiveness offering a lesson in how high-quality living space can be provided in high-density areas such as Tokyo.Dutch architect Martin van der Linden of YouTube channel One Minute Architecture hailed it as "a possibility for an alternative density of habitats"."It shows a possibility for an alternative density of habitats in the city as a collection of cell-like architectures or an architecture as the city," he said.Nor does it appear, in the Japanese architectural imagination, that there are any limits to what a dwelling can be Oliver WainwrightJapan is a breeding ground for experimental housing like Moriyama House, largely a symptom of the short lifespan of dwellings there, which averages out at 30 years in Tokyo.This can be pinned to the evolution of housing in Japan in the wake of world war two, and the country's rapid population growth in the 1960s and 1970s that led architects to develop innovative, sometimes wacky, solutions for living in hyperdense metropolises.Today, Moriyama House is arguably the most famous Japanese home in the world and one of the most notable examples of 21st-century Japanese architecture, epitomising the country's experimental and innovative approach to construction.The design rewrites the typical rules of privacy when it comes to housingThough much-loved by architects around the world, it was catapulted further into the spotlight in 2017 when it became a focal point of a landmark exhibition at London's Barbican Centre.The Japanese House Architecture and Life after 1945 centred around a 1:1 mock-up of the home in the gallery's central space. It was co-produced by the Japan Foundation to showcase the changes in Japanese domestic architecture since the end of the second world war.Read: Zaha Hadid's Bergisel Ski Jump was the most significant building of 2002Its curator Florence Ostend selected the house for the main exhibit with the belief that it is "one of the most important houses of the 21st century".In his review of the exhibition, critic Oliver Wainwright shared a similar sentiment, referring to it as "the most startling domestic vision of all"."Visitors will not need to fly to Tokyo to experience the most startling domestic vision of all, thanks to a 1:1 recreation of Ryue Nishizawa's seminal 2005 Moriyama House," he wrote."As Nishizawa puts it: 'life can't be contained within a single lot. People's sense of living expands beyond it, effectively erasing all borders.' Nor does it appear, in the Japanese architectural imagination, that there are any limits to what a dwelling can be."Did we get it right? Was Moriyama House by Ryue Nishizawa the most significant building completed in 2005? Let us know in the comments. We will be running a poll once all 25 buildings are revealed to determine the most significant building of the 21st century so far.This article is part of Dezeen's 21st-Century Architecture: 25 Years 25 Buildings series, which looks at the most significant architecture of the 21st century so far. For the series, we have selected the most influential building from each of the first 25 years of the century.The illustration is by Jack Bedford and the photography is by Edmund Sumner.21st-Century Architecture: 25 Years 25 Buildings2000: Tate Modern by Herzog & de Meuron2001: Gando Primary School by Dibdo Francis Kr2002: Bergisel Ski Jump by Zaha Hadid2003: Walt Disney Concert Hall by Frank Gehry2004: Quinta Monroy by Elemental2005: Moriyama House by Ryue NishizawaThis list will be updated as the series progresses.The post Ryue Nishizawa's Moriyama House was the most significant building of 2005 appeared first on Dezeen.
    0 Σχόλια ·0 Μοιράστηκε ·147 Views
  • Seven listening bars that are easy on the eyes and ears
    www.dezeen.com
    From a communist-era apartment block in Bucharest to a hot pink Australian bar created to emulate the inside of a jumbo speaker, our latest lookbook collects atmospheric listening bar interiors from around the world.Listening bars are having a moment. Native to Japan, these spaces first emerged during the late 1920s as intimate settings to meet, drink, and most importantly, listen to music together.Since then, the bars also known as hi-fi or audiophile bars have increasingly spread across the globe. Typically equipped with large sound systems and a wide selection of vinyl records, listening bars are popular for their unique musical offerings.This lookbook explores how international architects and designers have created the interiors of listening bars to respond to this well-loved model.This is the latest in our lookbooks series, which provides visual inspiration from Dezeen's archive. For more inspiration, see previous lookbooks featuring light-filled cottages, exposed concrete bedrooms and sculptural stone sinks.Top: photo by Ollie Tomlinson. Above: photo by Polly TootalKioku Bar, UK, by Pirajean LeesKioku Bar is a single-room sake bar on the ground floor of London's OWO hotel, attached to the Kioku restaurant upstairs. Local studio Pirajean Lees designed bothvenuesto capture head chef Endo Kazutoshi's recollections of living and working inJapanandSpain.Deep red dado and natural clay walls were paired with knobbly timber accents in the low-lit space. In one corner, a bespoke solid wood turntable was positioned for guests to play a selection of Japanese records from Endo's personal collection.Find out more about Kioku Bar Photo by Vlad PtruBar Ton, Romania, by Anda Zota and Muromuro StudioLocal designers Anda Zota and Muromuro Studio transformed a former music shop on the ground floor of a communist-era apartment block in Bucharest into a hi-fi bar.Bar Ton was created to blend its historical context with modern functionality. The designers maintained the open-plan space's marble mosaic floors and four concrete pillars, adding a timber unit between one pair to hold the main DJ booth. Birch plywood record shelves were inserted between the other two pillars."The overarching concept was about putting the music at the centre, both physically and conceptually," explained Zota.Find out more about Bar Ton Photo by Tim SalisburyJAM Record Bar, Australia, by Akin AtelierOwner Justin Hemmes and designer Kelvin Ho of Akin Atelier drew inspiration from Tokyo's many listening bars when creating the hot pink JAM Record Bar in Sydney.Distinct zones were organised around a central bar while Ho and his team chose specific materials to enhance the acoustic properties of the space from pink fibreglass insulation panels to sound-absorbing cork. Despite its small floor plan, Bar Ton houses a library of 15,000 records stacked in built-in plywood shelving."JAM is designed to feel like you are drinking a cocktail inside a giant speaker box but pink," said Ho.Find out more about JAM Record Bar Photo by Ill GanderUpstairs, USA, by Shane Davis and Francis HarrisThe founders of Brooklyn's "music-driven social space" Public Records added a listening bar and lounge to the upper floor of the creative venue.Undulated walls bounce music around the room from large subwoofer speakers, while cream leather banquettes and black ceramic and foam stools provide space for listeners to sit and choose from a curated selection of records and CDs.Find out more about Upstairs Photo by Rory GardinerGoodbye Horses, UK, by Leopold Banchini Architects"All the elements" of Swiss studio Leopold Banchini Architects' Goodbye Horses wine and listening bar were crafted from a single oak tree, including custom-made shelves that hold records.Located in northeast London's De Beauvoir neighbourhood, Goodbye Horses was designed to reference classic English pubs. The space features a hand-textured lime plaster ceiling and beaten earth flooring, chosen for the material's acoustical properties.Find out more about Goodbye Horses Photo by Nicholas WorleyAnima, Germany, by Grzywinski+PonsAmina is a restaurant and audiophile bar on the ground floor of the Locke at East Side Gallery hotel in Berlin, designed by New York architecture studioGrzywinski+Pons.Created as a dedicated space for music lovers, Amina features a rich material palette with subtle hues. Locally sourced bricks made from recycled sand and lime were used to fabricate built-in seating and planters, while large sandy-hued speakers were nestled on either side of homely white shelving stacked with records.Find out more about Anima Photo by Ollie TomlinsonSpace Talk, UK, by EBBA Architects and Charlotte TaylorLondon studio EBBA Architects collaborated with British designer Charlotte Taylor to create Space Talk, a hi-fi bar in the city's Clerkenwell neighbourhood.Divided into four distinct zones, each area of the bar features a different curation of sound and light to encourage visitors to move through the atmospheric space across an evening. Vintage furniture and objects were paired with acoustic panelling and woodblock columns, forming a distinctly eclectic interior.Find out more about Space Talk This is the latest in our lookbooks series, which provides visual inspiration from Dezeen's archive. For more inspiration, see previous lookbooks featuring light-filled cottages, exposed concrete bedrooms and sculptural stone sinks.The post Seven listening bars that are easy on the eyes and ears appeared first on Dezeen.
    0 Σχόλια ·0 Μοιράστηκε ·140 Views
  • Doom swaps blood for fine wine in this jokey remake
    www.creativebloq.com
    Doom: The Gallery Experience targets art history.
    0 Σχόλια ·0 Μοιράστηκε ·156 Views
  • Los Angeles Will Remain at High Risk of Fire Into Next Week
    www.wired.com
    The arrival of La Nia is starving California of rain, and more high Santa Ana winds could be on the way.
    0 Σχόλια ·0 Μοιράστηκε ·129 Views
  • Our 8 Favorite Indoor Air Quality Monitors Weve Tried (2025)
    www.wired.com
    These WIRED-tested indoor air-quality monitors have been teaching us things about our air quality we can never unsee.
    0 Σχόλια ·0 Μοιράστηκε ·137 Views
  • Jeff Bezos Blue Origin Rocket Launch Could Give SpaceX Some Competition
    www.nytimes.com
    If New Glenn lifts off overnight as planned, the Amazon founders rocket company will be on track to give Elon Musks SpaceX some genuine competition.
    0 Σχόλια ·0 Μοιράστηκε ·143 Views
  • Anthropics chief scientist on 5 ways agents will be even better in 2025
    www.technologyreview.com
    Agents are the hottest thing in tech right now. Top firms from Google DeepMind to OpenAI to Anthropic are racing to augment large language models with the ability to carry out tasks by themselves. Known as agentic AI in industry jargon, such systems have fast become the new target of Silicon Valley buzz. Everyone from Nvidia to Salesforce is talking about how they are going to upend the industry.We believe that, in 2025, we may see the first AI agents join the workforce and materially change the output of companies, Sam Altman claimed in a blog post last week.In the broadest sense, an agent is a software system that goes off and does something, often with minimal to zero supervision. The more complex that thing is, the smarter the agent needs to be. For many, large language models are now smart enough to power agents that can do a whole range of useful tasks for us, such as filling out forms, looking up a recipe and adding the ingredients to an online grocery basket, or using a search engine to do last-minute research before a meeting and producing a quick bullet-point summary.In October, Anthropic showed off one of the most advanced agents yet: an extension of its Claude large language model called computer use. As the name suggests, it lets you direct Claude to use a computer much as a person would, by moving a cursor, clicking buttons, and typing text. Instead of simply having a conversation with Claude, you can now ask it to carry out on-screen tasks for you.Anthropic notes that the feature is still cumbersome and error-prone. But it is alreadyavailable to a handful of testers, including third-party developers at companies such as DoorDash, Canva, and Asana.Computer use is a glimpse of whats to come for agents. To learn whats coming next, MIT Technology Review talked to Anthropics cofounder and chief scientist Jared Kaplan. Here are five ways that agents are going to get even better in 2025.(Kaplans answers have been lightly edited for length and clarity.)1/ Agents will get better at using toolsI think there are two axes for thinking about what AI is capable of. One is a question of how complex the task is that a system can do. And as AI systems get smarter, theyre getting better in that direction. But another direction thats very relevant is what kinds of environments or tools the AI can use.So, like, if you go back almost 10 years now to [DeepMinds Go-playing model] AlphaGo, we had AI systems that were superhuman in terms of how well they could play board games. But if all you can work with is a board game, then thats a very restrictive environment. Its not actually useful, even if its very smart. With text models, and then multimodal models, and now computer useand perhaps in the future with roboticsyoure moving toward bringing AI into different situations and tasks, and making it useful.We were excited about computer use basically for that reason. Until recently, with large language models, its been necessary to give them a very specific prompt, give them very specific tools, and then theyre restricted to a specific kind of environment. What I see is that computer use will probably improve quickly in terms of how well models can do different tasks and more complex tasks. And also to realize when theyve made mistakes, or realize when theres a high-stakes question and it needs to ask the user for feedback.2/ Agents will understand context Claude needs to learn enough about your particular situation and the constraints that you operate under to be useful. Things like what particular role youre in, what styles of writing or what needs you and your organization have.ANTHROPICI think that well see improvements there where Claude will be able to search through things like your documents, your Slack, etc., and really learn whats useful for you. Thats underemphasized a bit with agents. Its necessary for systems to be not only useful but also safe, doing what you expected.Another thing is that a lot of tasks wont require Claude to do much reasoning. You dont need to sit and think for hours before opening Google Docs or something. And so I think that a lot of what well see is not just more reasoning but the application of reasoning when its really useful and important, but also not wasting time when its not necessary.3/ Agents will make coding assistants betterWe wanted to get a very initial beta of computer use out to developers to get feedback while the system was relatively primitive. But as these systems get better, they might be more widely used and really collaborate with you on different activities.I think DoorDash, the Browser Company, and Canva are all experimenting with, like, different kinds of browser interactions and designing them with the help of AI.My expectation is that well also see further improvements to coding assistants. Thats something thats been very exciting for developers. Theres just a ton of interest in using Claude 3.5 for coding, where its not just autocomplete like it was a couple of years ago. Its really understanding whats wrong with code, debugging itrunning the code, seeing what happens, and fixing it.4/ Agents will need to be made safeWe founded Anthropic because we expected AI to progress very quickly and [thought] that, inevitably, safety concerns were going to be relevant. And I think thats just going to become more and more visceral this year, because I think these agents are going to become more and more integrated into the work we do. We need to be ready for the challenges, like prompt injection.[Prompt injection is an attack in which a malicious prompt is passed to a large language model in ways that its developers did not foresee or intend. One way to do this is to add the prompt to websites that models might visit.]Prompt injection is probably one of the No.1 things were thinking about in terms of, like, broader usage of agents. I think its especially important for computer use, and its something were working on very actively, because if computer use is deployed at large scale, then there could be, like, pernicious websites or something that try to convince Claude to do something that it shouldnt do.And with more advanced models, theres just more risk. We have a robust scaling policy where, as AI systems become sufficiently capable, we feel like we need to be able to really prevent them from being misused. For example, if they could help terroriststhat kind of thing.So Im really excited about how AI will be usefulits actually also accelerating us a lot internally at Anthropic, with people using Claude in all kinds of ways, especially with coding. But, yeah, therell be a lot of challenges as well. Itll be an interesting year.
    0 Σχόλια ·0 Μοιράστηκε ·153 Views
  • When the U.S. Surgeon General Shocked Americans by Announcing That Smoking Kills
    www.smithsonianmag.com
    U.S. Surgeon General Luther Terry shocked Americans by announcing that cigarette smoking is a major cause of lung cancer, throat cancer and chronic bronchitis. Bettmann via Getty ImagesSixty years ago, about 70 million Americans smoked tobacco. An estimated 42 percent of adults identified themselves as smokers in 1965, and advertisements for cigarettes were impossible to avoid. Tobacco products were stylish and healthy, manufacturers insisted, with Camel claiming its cigarettes dont get your wind and Old Gold saying its were fresh as mountain air.So it came as a shock to the American public when, on January 11, 1964, their surgeon general appeared on television saying that smoking tobacco leads to disease and death.Luther Terry, appointed surgeon general by President John F. Kennedy in 1961, had just reviewed the results of a 14-month study by a committee of top-notch medical scientists. Their conclusions, presented in a 387-page report, were that cigarette smoking is a major cause of lung cancer, throat cancer and chronic bronchitis.In short, the committee says if you smoke cigarettes, you increase your chances of dying early, CBS News reported at the time. The sooner you start, the more you smoke, the more you inhale, the worse your chances are.Smoking: "A health hazard of sufficient importance"Watch on The ten-man committee, made up of five smokers and five non-smokers, took questions from reporters at a taped press conference announcing their findings. They pointed out that 41,000 Americans died annually from lung cancermore than the annual number of automobile accident fatalities, which was about 38,000. Overall, about 752,000 Americans died of cigarette-associated causes in 1962 alone, they said. Terry stated clearly that hed advise anybody smoking cigarettes to quit.This connection between cigarettes and cancer wasnt exactly new: Doctors in America had been paying attention to smokings pulmonary effects for decades. As lung surgeon Alton Ochsner wrote in 1939, Inhaled smoke, constantly repeated over a long period of time, undoubtedly is a source of chronic irritation to the bronchial mucosa. Researchers also noted an increase in lung cancer cases, but the incline was commonly attributed to other types of inhaled toxins, like automobile exhaust, tar fumes and factory smog.As far back as the 1920s, researchers had recorded the smoking habits of people who developed lung cancer and compared them to cancer-free people. They found strong associations between smoking and cancer, so researchers followed up, beginning more studies in the early 1950sthe research that eventually led to the surgeon generals report. A warning label on a package of Marlboro cigarettes Wikimedia Commons under CC BY-SA 3.01964 marked the governments arrival to the non-smoking movement. A year after the top doctors announcement, American smokers arrived in stores to find their cigarette boxes stamped with a mandatory warning label: Caution: Cigarette smoking may be hazardous to your health.That weakly worded caution strengthened over the years as public health officials grew firmer on their stance that tobacco causes cancer and many other health problems. Now, the Food and Drug Administration prescribes 11 different warning labels for tobacco products, complemented by deliberately frightening images. Smoking reduces blood flow to the limbs, which can require amputation, the labels declare. Smoking causes head and neck cancer.Though it took decades for the number of American smokers to significantly decreasea trend challenged by a 1980s uptick in youth smokingsmoking rates have fallen each year since the surgeon generals first warning. Between 1941 and 1974, surveyor Gallup annually asked Americans if theyd smoked a cigarette in the past week. At least four in ten responded yes in each survey. As of 2024, that number is down to one.Get the latest stories in your inbox every weekday.Filed Under: Advertisements, American History, Cancer, Disease, Disease and Illnesses, Health, On This Day in History, Tobacco
    0 Σχόλια ·0 Μοιράστηκε ·131 Views
  • Anthropics chief scientist on 5 ways agents will be even better in 2025
    www.technologyreview.com
    Agents are the hottest thing in tech right now. Top firms from Google DeepMind to OpenAI to Anthropic are racing to augment large language models with the ability to carry out tasks by themselves. Known as agentic AI in industry jargon, such systems have fast become the new target of Silicon Valley buzz. Everyone from Nvidia to Salesforce is talking about how they are going to upend the industry. We believe that, in 2025, we may see the first AI agents join the workforce and materially change the output of companies, Sam Altman claimed in a blog post last week. In the broadest sense, an agent is a software system that goes off and does something, often with minimal to zero supervision. The more complex that thing is, the smarter the agent needs to be. For many, large language models are now smart enough to power agents that can do a whole range of useful tasks for us, such as filling out forms, looking up a recipe and adding the ingredients to an online grocery basket, or using a search engine to do last-minute research before a meeting and producing a quick bullet-point summary. In October, Anthropic showed off one of the most advanced agents yet: an extension of its Claude large language model called computer use. As the name suggests, it lets you direct Claude to use a computer much as a person would, by moving a cursor, clicking buttons, and typing text. Instead of simply having a conversation with Claude, you can now ask it to carry out on-screen tasks for you. Anthropic notes that the feature is still cumbersome and error-prone. But it is alreadyavailable to a handful of testers, including third-party developers at companies such as DoorDash, Canva, and Asana. Computer use is a glimpse of whats to come for agents. To learn whats coming next, MIT Technology Review talked to Anthropics cofounder and chief scientist Jared Kaplan. Here are five ways that agents are going to get even better in 2025. (Kaplans answers have been lightly edited for length and clarity.) 1/ Agents will get better at using tools I think there are two axes for thinking about what AI is capable of. One is a question of how complex the task is that a system can do. And as AI systems get smarter, theyre getting better in that direction. But another direction thats very relevant is what kinds of environments or tools the AI can use. So, like, if you go back almost 10 years now to [DeepMinds Go-playing model] AlphaGo, we had AI systems that were superhuman in terms of how well they could play board games. But if all you can work with is a board game, then thats a very restrictive environment. Its not actually useful, even if its very smart. With text models, and then multimodal models, and now computer useand perhaps in the future with roboticsyoure moving toward bringing AI into different situations and tasks, and making it useful. We were excited about computer use basically for that reason. Until recently, with large language models, its been necessary to give them a very specific prompt, give them very specific tools, and then theyre restricted to a specific kind of environment. What I see is that computer use will probably improve quickly in terms of how well models can do different tasks and more complex tasks. And also to realize when theyve made mistakes, or realize when theres a high-stakes question and it needs to ask the user for feedback. 2/ Agents will understand context Claude needs to learn enough about your particular situation and the constraints that you operate under to be useful. Things like what particular role youre in, what styles of writing or what needs you and your organization have. ANTHROPIC I think that well see improvements there where Claude will be able to search through things like your documents, your Slack, etc., and really learn whats useful for you. Thats underemphasized a bit with agents. Its necessary for systems to be not only useful but also safe, doing what you expected. Another thing is that a lot of tasks wont require Claude to do much reasoning. You dont need to sit and think for hours before opening Google Docs or something. And so I think that a lot of what well see is not just more reasoning but the application of reasoning when its really useful and important, but also not wasting time when its not necessary. 3/ Agents will make coding assistants better We wanted to get a very initial beta of computer use out to developers to get feedback while the system was relatively primitive. But as these systems get better, they might be more widely used and really collaborate with you on different activities. I think DoorDash, the Browser Company, and Canva are all experimenting with, like, different kinds of browser interactions and designing them with the help of AI. My expectation is that well also see further improvements to coding assistants. Thats something thats been very exciting for developers. Theres just a ton of interest in using Claude 3.5 for coding, where its not just autocomplete like it was a couple of years ago. Its really understanding whats wrong with code, debugging itrunning the code, seeing what happens, and fixing it. 4/ Agents will need to be made safe We founded Anthropic because we expected AI to progress very quickly and [thought] that, inevitably, safety concerns were going to be relevant. And I think thats just going to become more and more visceral this year, because I think these agents are going to become more and more integrated into the work we do. We need to be ready for the challenges, like prompt injection. [Prompt injection is an attack in which a malicious prompt is passed to a large language model in ways that its developers did not foresee or intend. One way to do this is to add the prompt to websites that models might visit.] Prompt injection is probably one of the No.1 things were thinking about in terms of, like, broader usage of agents. I think its especially important for computer use, and its something were working on very actively, because if computer use is deployed at large scale, then there could be, like, pernicious websites or something that try to convince Claude to do something that it shouldnt do. And with more advanced models, theres just more risk. We have a robust scaling policy where, as AI systems become sufficiently capable, we feel like we need to be able to really prevent them from being misused. For example, if they could help terroriststhat kind of thing. So Im really excited about how AI will be usefulits actually also accelerating us a lot internally at Anthropic, with people using Claude in all kinds of ways, especially with coding. But, yeah, therell be a lot of challenges as well. Itll be an interesting year.
    0 Σχόλια ·0 Μοιράστηκε ·132 Views