A new Friends Tab will feature posts from a users friends and relatives, which was the original mission of the app.

MacworldAt a glanceExpert's RatingProsAffordableBright screenHigh-quality design and workmanshipStrong performanceFast 5G and WLANImpressive battery lifeConsNo Apple IntelligenceGlossy displayHeavier and thicker than iPad Pro and AirLow USB-C data transfer rateLimited keyboard supportNo Stage ManagerOur Verdict Apples most affordable iPad models, such as the 9th- and 10th-gen iPads, have always been bestsellers. Thanks to perfectly adequate performance and sufficient storage space, this new model will also be a success. The lack of support for Apple Intelligence is a drawback, but there are plenty of alternatives from Google, OpenAI, and others that you can use to fill the AI gap in your life.Price When ReviewedThis value will show the geolocated pricing text for product undefinedBest Pricing TodayPrice When Reviewed399 EuroBest Prices Today: Apple iPad (A16)RetailerPrice385View Deal389View Deal389View Deal399View Deal380.9View Deal377.83View Deal377,90 View Deal380,90 View Deal380,90 View Deal389,00 View Deal389,99 View Deal389,99 View Deal410,00 View DealPrice comparison from over 24,000 stores worldwideView more pricesProductPricePrice comparison from BackmarketThere are good reasons why Apples cheapest tablets, such as the 9th- and 10th-gen iPads, have been big sellers. In theory, Apples standard iPad models are designed primarily for educational institutions, but in reality, they offer everything most users want. Most of the time a basic iPad is completely sufficient.The last of these standard iPads, which came out in 2022, sold quite well despite its initially high price of $449. (It later dropped to a more palatable $349.) This model was the first in the line to have a more modern design, featuring a USB-C port and a Touch ID sensor in the power button, which brought it closer to the top models. But there were compromises too: buyers had to make do with an older chip and mediocre features.With the new 11th-gen iPad modelalso known as the iPad (A16)Apple has again cut corners compared to the iPad Air and Pro. For example, Apple has selected the low-cost A16 chip from the iPhone 15, which does not support Apple Intelligence but more on that later.iPad (A16) design & build quality: Tried and testedCar reviewers love to discuss quality feel, and in a similar way I got an intangible impression of excellence from the iPad. It just feels good. The casing is similar to that of the current iPad Air models, but a little thicker. This is due to the simpler unlaminated display, which takes up more space. The current iPad Pro 11-inch is only 5.3mm thick, whereas the iPad Air is 6.1mm and the iPad (A16) is 7mm. The weight is also slightly higher: this device is 17g heavier than the current 11-inch iPad Air.The color scheme underlines the fact that the target audience includes younger users. As before, you get a choice of blue, pink, yellow, and silver. Our sample was yellow: a very classy-looking shade of golden yellow. The colors are much stronger than those of the iPad Air models, which are more discreet and grown-up. Business users can choose neutral silver.Eugen WegmanniPad (A16) display: Sharp and (too) shinyThe resolution and color reproduction of the iPads screen are good, and at 500 nits its bright enough for working outdoors. Although it doesnt support the P3 color space offered by the Pro and Air models, this is irrelevant for most users. Its also worth noting that while Apple is referring to an 11-inch display for the first time on this line (the 10th-gen iPad is referred to on the companys website as having a 10.9-inch screen), the size and resolution are identical to its predecessor.The displays main weak point is its highly reflective surface. Although the screens of an iPad Air or mini are also reflective, the reflections are much stronger on the basic iPads. This quickly becomes annoying in unfavorable lighting conditions: if, for example, youre sitting next to a window or a lamp. However, these reflections dont bother all buyers and shouldnt be overstated. A truly matte display is only available on the iPad Pro, and this is not only expensive but also more susceptible to scratches.iPad (A16) specs: Useful upgradesWhen Apple announced this iPad, the big surprise was that it doesnt support Apple Intelligence. Its the only current iPadOS or iOS device not to be compatible with Apples AI platform.The iPad doesnt quite have the specs to run Apple Intelligence, but it does have some useful upgrades. The A14 chip in the 2022 iPad has been bumped to a faster A16 Bionic, which is a solid leap: the A14 was introduced with the iPhone 12, and the A16 in the iPhone 15. The available RAM also increases from 4 to 6GB.I suspect that the extra RAM in particular could be very useful when iPadOS 19 is released. The new operating system will have many new features and a new interface, and 4GB could quickly become insufficient. With the new iPad, you can be sure you wont have any performance problems for the next few years, even if you have to manage without Apple Intelligence. Youll also have to do without some functions such as Stage Manager.The storage capacity is also a real plus point: even the basic model offers 128GB, which is sufficient for most users. Versions with 256 and 512GB are also available. With the previous model, there was only a choice between 64GB (too little) and 256GB (too expensive).The audio quality of the stereo speakers, which are positioned at the top and bottom, is good. As in most areas, however, the iPad cannot quite keep up with the iPad Air and Pro.The volume is completely sufficient for YouTube or Apple TV and the sound quality is acceptable and significantly better than the 9th-gen iPad and earlier. Two microphones ensure good sound quality for Teams and similar.Eugen WegmanniPad (A16) features: Why no Apple Intelligence?Theres been a lot of speculation as to why Apple has left out its generative AI tools, but the company has so far declined to elaborate. I assume there were two reasons for this: Cupertino wanted to save as much money as possible on this model and not make the iPad too attractive. Its predecessor was one of the best-selling tablets on the market and certainly cannibalized the sales of the more profitable iPad Air.The A16 looks to be Apples new budget chip; its even being produced by TSMC for the first time in the US in Arizona, which means it avoids tariffs. Apple production costs are not known, but by choosing the A16 instead of something newer and more powerful, it should be possible to save several dollars per unit.In any case, the new iPads performance is perfectly adequate for education and business users, and some schools may actually prefer for their pupils not to have access to Apples AI text tools. Price is key in the education market; school budgets are tight and Apples iPad is in strong competition with Googles Chromebooks.The pill is pretty easy to swallow in any case because Apple Intelligences development is behind schedule and well probably have to wait until iPadOS 19 for the promised new AI version of Siri. Alternatives such as ChatGPT, Google Gemini and Microsoft Copilot are available as apps.iPad (A16) performance: Fast enoughAs already mentioned, the iPad has the same chip as the iPhone 15, making it fast enough for most current apps and tasks such as image editing and video editing with iMovie. However, it lacks the multiprocessor performance of an M3 chip, which could become an issue if youre keen on games and more demanding video projects. In everyday use and when browsing the web, its primarily the single-CPU performance thats required, and youll struggle to notice a difference between this device and something with a faster chip.Surprisingly, the new iPad is even compatible with Final Cut Pro, whereas models with the M1 chip are not. However, some functions of the editing software are not supported.Its worth pointing out that the A16 chip in the iPad is slightly weaker than that of the iPhone 15. Here, we get a 5-core CPU and a 4-core GPU. By contrast, the iPhone 15 gets an A16 with a 6-core CPU and a 5-core GPU, so were losing a core in each department. Only the Neural Engine is the same, offering 16 cores in both iterations. This version of the A16 is presumably based on product binning, deliberately selecting chips with fewer functional cores at the production stage.In our speed benchmarks, single-core performanceimportant for apps and the webwas outstanding. The iPad Air M3 was just under 10 percent faster in the Speedometer and JetStream benchmarks. The iPad Air scored 33 points in Speedometer, while the iPad (A16) scored 29.7 points.The Geekbench 5 benchmark illustrates the iPads enhanced performance.FoundryTesting did however reveal some weaknesses in graphics performance which is no wonder, with only four cores in action. Whereas the iPad Air (M3) scored an excellent 46,007 points in the Geekbench 6 graphics benchmark, the iPad (A16) achieved less than half that, with 20,123 points. The iPhone 15 scored 22,589.In the demanding Wildlife Extreme test, the iPads score of 2,675 was significantly lower than the iPad Airs 6,446 and the iPhone 15s 3,013.More positively, overheating was not a problem for the iPad. In the 20-minute test using APSI Bench, performance was only reduced to 93.94 percent. With the much faster iPad Air (M3) the figure was 86.57 per cent.The iPad can handle high continuous load.IDGiPad (A16) accessories & connectivity: Some limitationsLike its predecessor, the iPad (A16) only supports the simpler Apple Pencil with USB-C and the first-gen Pencil. This could be because many schools have bought these input devices. However, this is annoying for owners who already have a second-gen Pencil or even a Pencil Pro. These accessories are simply not recognized.You can see the Smart Connector interface on the side, but its positioned differently from other iPads. Of Apples own iPad cases, only the Smart Folio cases and the bulky Magic Keyboard Folio are compatible. The latter is not cheap and offers a completely different concept to the elegant Magic Keyboards for iPad Pro and Air.Dont forget that you can also easily connect the iPad to a USB-C monitor. If you already have a mouse and keyboard attached to the monitor, you can start working immediately. Unfortunately, the display does not support the Stage Manager function, which is more convenient for this working mode. I felt the absence of a headphone output, but this has been missing since the 9th-gen iPad. On the plus side theres a USB-C port which, with the addition of a few inexpensive adapters and hubs, gives you a wide range of options such as SD cards, HDMI and audio.The USB-C port on the bottom is unfortunately slower than on other iPads. While the iPad Pro and Air can transfer data to an external SSD at high speed, this iPad offers only USB 2.0 speeds. When copying to an external SSD from LaCie, we measured 37 MB/s for a video file, and performance can be significantly slower when youre working with many small files. In contrast, the current iPad Air achieved just over 700 MB/s in our test.Apple has cut back a little on Wi-Fi support: the current iPad Air and Pro support the fast Wi-Fi 6E, while the iPad (A16) still has to make do with Wi-Fi 6. However, this is not a problem for most consumers who wont have a router with Wi-Fi 6E anyway.The models with mobile network support support 5G via eSIM. A physical SIM card is no longer supported.Eugen WegmanniPad (A16) battery life: Impressively long-lastingApple promises up to 10 hours of surfing the web on WiFi or watching video. Thats the same estimate as on the previous generation and every other iPad generation.Of course, we didnt take Apples word for it and ran our own tests. We let a TV show run on a continuous loop with the screen set to maximum brightness and True Tone disabled. (Apple uses a screen brightness of 50 percent for its measurements.) The 7,544mAh battery lasted for six hours and ten minutes, just like the 10th-gen iPad. It also lasted seven hours in our web surfing test at 50 percent brightness. These are very useful numbers overall. The iPad could probably only be a little more power-efficient if it was equipped with an OLED display.iPad (A16) camera: Little changeAs with the previous model, the front camera is positioned on the long edge, which is the optimal placement for video calls. Technically, nothing has changed here: its a 12MP camera with support for Center Stage, a brilliant feature that uses machine-learning technology to pan and zoom in and get the perfect shot during a FaceTime call or other videoconference.The rear camera is still 12MP. Light intensity and resolution are also moderate, with up to 5x digital zoom. Unfortunately, theres no flash, a function that is reserved for the Pro models.Video recordings are still possible at up to 4K and 60fps, and video image stabilization is also available.Should you buy the iPad (A16)?For $349/329, you get a first-class iPad with good performance and sufficient storage space. However, graphics performance is a weak point; there are better models for gamers. The iPad Air and Pro are superior in many areas and offer better displays, but they are very expensive. More casual users should also consider the elegant iPad mini, which offers a smaller but better display and higher performance and is often overlooked.The lack of support for Apple Intelligence is regrettable, but it will probably be another year before the function is fully developed. And in any case, ChatGPT and Gemini are available as standalone apps on the iPad so youre not missing out completely.

MacworldApple has been holding back from the foldables market for so long now that at times it seemed like it would never release a folding iPhone. But this month has seen a succession of more and more plausible reports that suggest the iPhone Fold project is finally happening, even if a launch remains a couple of years away.Near the start of March, the respected analyst Ming-Chi Kuo outlined his expectations for the deviceultra-thin, folding in the book style rather than a clamshell, with a crease-free screen and a $2,000-plus price tag. Then Mark Gurman revealed that the iPhone 17 Air expected to launch this fall will act as a trial run leading to major iPhone redesign plans, including the ability to fold, if its successful. Finally, a Barclays analyst narrowed pricing expectations to around $2,300.But the most interesting rumor arrived last week. We were back to Kuo, who this time predicted that the iPhone Fold would have a hinge made of LiquidMetal. (Not literally metal in its liquid form, but the marketing term for a material that uses a special casting process to produce exceptionally strong, hard, and precise metal parts.) This seemed especially plausible because Apple purchased worldwide exclusive rights to use material developed by a related company named LiquidMetal 15 years ago.The rumor train shows no sign of stopping, and the well-known and prolific Weibo leaker Instant Digital posted a message Thursday (via MacRumors) offering further details of the hinge mechanism Apple is working on for its first foldable iPhone. The hinge will be made of amorphous alloy (metallic glass), the leaker said, metallic glass being another term for liquid metal, before adding explanations of why such a material would be chosen.Advantages, Instant began. Amorphous particle structure, better resistance to bending, deformation, and dents, strength and hardness are 2.5 times higher than titanium alloy, and glossiness is also better (basically close to the high-end feeling of stainless steel that you like). All very appealing. But in response to a question in the comments about whether screen creases (one of the main issues with current folding phones) can disappear with the use of this material, the leaker simply says impossible. So well have to wait to see how Apple solves that, which is said to be a priority.We dont currently expect the iPhone Fold to launch until 2027. For all the latest news and rumors, bookmark our regularly updated folding iPhone superguide.

The Group of 7 (G7) nations are falling behind in key tech economic indicators such as high-tech exports, the number of software developers, and AI-related patent filings, endangering their future competitiveness, according to new study by London-based workforce consultancy SThree and the Center for Economics and Business Research (Cebr).In fact, not one G7 nation made it into the top 10 a clear warning sign for the future, the study warned. (The G7 consists of the US, UK, Canada, France, Germany, Italy, Japan, and the European Union as a non-enumerated member.)Once the global epicenter for innovation, these countries are now facing stiff competition from emerging tech hubs, said SThree CEO Timo Lehne. The challenge is no longer simply about maintaining their position; its about ensuring they lead the charge in fostering innovation and nurturing the businesses that will drive the future of global technology.Without focusing on innovation and future industries, tech leadership by the G7 is no longer guaranteed, Lehne said.Tech advances are reshaping the global economy, with industries such as AI, automation, and clean energy relying on a STEM-skilled workforce. As a result, countries investing in science, technology, engineering and math (STEM) education and training will drive growth; those that dont may fall behind, according to the report.The G7 has already seen the effects of reduced competitiveness this year, with the USs Magnificent Seven (Apple, Microsoft, Amazon, Nvidia, Alphabet, Meta, and Tesla)losing $1.5 trillion in market value since the start of 2025. The companies saw similar losses in just a few days last year as well.According to SThree, the US has slipped several spots and is now behind the UK and Canada in tech competitiveness, while Singapore, Ireland, and Australia all secured top-10 spots based on STEM skills and training.Asian countries occupied the top spots in the Foundational Education Pillar, with Singapore on top, followed by Japan and South Korea. Estonia was the top-scoring European nation at No. 4.SThree and the Centre for Economics and Business Research Singapores success can be attributed to its focus on services, R&D, and innovation, according to industry observers. The nations government has an Economic Development Board (EDB), which works to attract and grow industries, and it has shifted its focus from low-cost manufacturing to high-value sectors such as aerospace and semiconductors. It has also become a hub for digital technologies, with many companies relocating their headquarters from Hong Kong due to Chinas influence over that territory, which it reclaimed in 1997.Education and research institutes in Singapore focus on developing a skilled workforce in tech fields like AI, while companies benefit from funded research partnerships. Singapore has also nurtured a number of tech unicorns including Lazada, Grab, and Ninja Van, promotes fintech through annual events, and has easy work visa access.China was not included in the study, as its data was deemed not reliable yet researchers noted that not including China is arguably the biggest omission in this years index. From what we know, Chinas STEM ecosystem is developing very quickly. It boasts 63 of the top 500 research institutions, it is increasingly seen as a research superpower that is competing with the likes of the United States and Europe, and is investing heavily in R&D3.Switzerland and Sweden got top marks for STEM skills, while Denmark passed Sweden for second place for Life Sciences, according to the study. Finland and the Republic of Korea saw improved scores on engineering skills, coming in first and second, respectively.Although the UK and US refused to sign last months European Union AI agreement regulating the technology, each ranks 11th and 16th, respectively, for AI patents, with Korea, Japan, and Singapore at the top. The lack of competitiveness in the G7 was felt when US tech giants lost $1.13 trillion in market value, affecting companies like Germanys Infineon and Japans SoftBank, the study said, pointing to losses in 2024. In overall tech rankings, Singapore, Ireland, and Australia lead in fostering tech innovation, surpassing all G7 nations.The importance of investing in STEMSingapore has skyrocketed in tech innovation and exports for a myriad of reasons, not the least of which is because STEM skills can boost critical thinking and problem-solving across any role. Cebr and SThree used 26 indicators in areas like education, workforce integration, industry opportunities, and innovation to develop their index ranking 35 countries based on STEM skills. Success depends on collaboration between governments, businesses, and education to build a skilled STEM workforce.A growing number of organizations are dropping traditional college degree requirements in favor of skills gained through alternative methods. Large companies, including Boeing, Walmart, and IBM, have signed on to varying skills-based employment projects, such as theRework America Alliance,the Business RoundtablesMultiple Pathways program,and the campaign toTear the Paper Ceiling,pledging to implement skills-based practices, according to McKinsey & Co.So far, theyve removed degree requirements from certain job postings and have worked with other organizations to help workers progress from lower- to higher-wage jobs, McKinsey said ina November report.Skills-based hiring helps companies find and attract a broader pool of candidates better suited to fill positions long term, and it opens up opportunities to non-traditional candidates, including women and minorities, according to McKinsey.

MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand whats coming next. You can read more from the series here.With the recent news that the Atlantics editor in chief was accidentally added to a group Signal chat for American leaders planning a bombing in Yemen, many people are wondering: What is Signal? Is it secure? If government officials arent supposed to use it for military planning, does that mean I shouldnt use it either?The answer is: Yes, you should use Signal, but government officials having top-secret conversations shouldnt use Signal.Read on to find out why.What is Signal?Signal is an app you can install on your iPhone or Android phone, or on your computer. It lets you send secure texts, images, and phone or video chats with other people or groups of people, just like iMessage, Google Messages, WhatsApp, and other chat apps.Installing Signal is a two-minute processagain, its designed to work just like other popular texting apps.Why is it a problem for government officials to use Signal?Signal is very secureas well see below, its the best option out there for having private conversations with your friends on your cell phone.But you shouldnt use it if you have a legal obligation to preserve your messages, such as while doing government business, because Signal prioritizes privacy over ability to preserve data. Its designed to securely delete data when youre done with it, not to keep it. This makes it uniquely unsuited for following public record laws.You also shouldnt use it if your phone might be a target of sophisticated hackers, because Signal can only do its job if the phone it is running on is secure. If your phone has been hacked, then the hacker can read your messages regardless of what software you are running.This is why you shouldnt use Signal to discuss classified material or military plans. For military communication your civilian phone is always considered hacked by adversaries, so you should instead use communication equipment that is saferequipment that is physically guarded and designed to do only one job, making it harder to hack.What about everyone else?Signal is designed from bottom to top as a very private space for conversation. Cryptographers are very sure that as long as your phone is otherwise secure, no one can read your messages.Why should you want that? Because private spaces for conversation are very important. In the US, the First Amendment recognizes, in the right to freedom of assembly, that we all need private conversations among our own selected groups in order to function.And you dont need the First Amendment to tell you that. You know, just like everyone else, that you can have important conversations in your living room, bedroom, church coffee hour, or meeting hall that you could never have on a public stage. Signal gives us the digital equivalent of thatits a space where we can talk, among groups of our choice, about the private things that matter to us, free of corporate or government surveillance. Our mental health and social functioning require that.So if youre not legally required to record your conversations, and not planning secret military operations, go ahead and use Signalyou deserve the privacy.How do we know Signal is secure?People often give up on finding digital privacy and end up censoring themselves out of caution. So are there really private ways to talk on our phones, or should we just assume that everything is being read anyway?The good news is: For most of us who arent individually targeted by hackers, we really can still have private conversations.Signal is designed to ensure that if you know your phone and the phones of other people in your group havent been hacked (more on that later), you dont have to trust anything else. It uses many techniques from the cryptography community to make that possible.Most important and well-known is end-to-end encryption, which means that messages can be read only on the devices involved in the conversation and not by servers passing the messages back and forth.But Signal uses other techniques to keep your messages private and safe as well. For example, it goes to great lengths to make it hard for the Signal server itself to know who else you are talking to (a feature known as sealed sender), or for an attacker who records traffic between phones to later decrypt the traffic by seizing one of the phones (perfect forward secrecy).These are only a few of many security properties built into the protocol, which is well enough designed and vetted for other messaging apps, such as WhatsApp and Google Messages, to use the same one.Signal is also designed so we dont have to trust the people who make it. The source code for the app is available online and, because of its popularity as a security tool, is frequently audited by experts.And even though its security does not rely on our trust in the publisher,it does come from a respected source: the Signal Technology Foundation, a nonprofit whose mission is to protect free expression and enable secure global communication through open-source privacy technology. The app itself, and the foundation, grew out of a community of prominent privacy advocates. The foundation was started by Moxie Marlinspike, a cryptographer and longtime advocate of secure private communication, and Brian Acton, a cofounder of WhatsApp.Why do people use Signal over other text apps? Are other ones secure?Many apps offer end-to-end encryption, and its not a bad idea to use them for a measure of privacy. But Signal is a gold standard for private communication because it is secure by default: Unless you add someone you didnt mean to, its very hard for a chat to accidentally become less secure than you intended.Thats not necessarily the case for other apps. For example, iMessage conversations are sometimes end-to-end encrypted, but only if your chat has blue bubbles, and they arent encrypted in iCloud backups by default. Google Messages are sometimes end-to-end encrypted, but only if the chat shows a lock icon. WhatsApp is end-to-end encrypted but logs your activity, including how you interact with others using our Services.Signal is careful not to record who you are talking with, to offer ways to reliably delete messages, and to keep messages secure even in online phone backups. This focus demonstrates the benefits of an app coming from a nonprofit focused on privacy rather than a company that sees security as a nice to have feature alongside other goals.(Conversely, and as a warning, using Signal makes it rather easier to accidentally lose messages! Again, it is not a good choice if you are legally required to record your communication.)Applications like WhatsApp, iMessage, and Google Messages do offer end-to-end encryption and can offer much better security than nothing. The worst option of all is regular SMS text messages (green bubbles on iOS)those are sent unencrypted and are likely collected by mass government surveillance.Wait, how do I know that my phone is secure?Signal is an excellent choice for privacy if you know that the phones of everyone youre talking with are secure. But how do you know that? Its easy to give up on a feeling of privacy if you never feel good about trusting your phone anyway.One good place to start for most of us is simply to make sure your phone is up to date. Governments often do have ways of hacking phones, but hacking up-to-date phones is expensive and risky and reserved for high-value targets. For most people, simply having your software up to date will remove you from a category that hackers target.If youre a potential target of sophisticated hacking, then dont stop there. Youll need extra security measures, andguides from the Freedom of the Press Foundation and the Electronic Frontier Foundation are a good place to start.But you dont have to be a high-value target to value privacy. The rest of us can do our part to re-create that private living room, bedroom, church, or meeting hall simply by using an up-to-date phone with an app that respects our privacy.Jack Cushman is a fellow of the Berkman Klein Center for Internet and Society and directs the Library Innovation Lab at Harvard Law School Library. He is an appellate lawyer, computer programmer, and former board member of the ACLU of Massachusetts.

The AI firm Anthropic has developed a way to peer inside a large language model and watch what it does as it comes up with a response, revealing key new insights into how the technology works. The takeaway: LLMs are even stranger than we thought.The Anthropic team was surprised by some of the counterintuitive workarounds that large language models appear to use to complete sentences, solve simple math problems, suppress hallucinations, and more, says Joshua Batson, a research scientist at the company.Its no secret that large language models work in mysterious ways. Fewif anymass-market technologies have ever been so little understood. That makes figuring out what makes them tick one of the biggest open challenges in science.But its not just about curiosity. Shedding some light on how these models work would expose their weaknesses, revealing why they make stuff up and can be tricked into going off the rails. It would help resolve deep disputes about exactly what these models can and cant do. And it would show how trustworthy (or not) they really are.Batson and his colleagues describe their new work in two reports published today. The first presents Anthropics use of a technique called circuit tracing, which lets researchers track the decision-making processes inside a large language model step by step. Anthropic used circuit tracing to watch its LLM Claude 3.5 Haiku carry out various tasks. The second (titled On the Biology of a Large Language Model) details what the team discovered when it looked at 10 tasks in particular.I think this is really cool work, says Jack Merullo, who studies large language models at Brown University in Providence, Rhode Island, and was not involved in the research. Its a really nice step forward in terms of methods.Circuit tracing is not itself new. Last year Merullo and his colleagues analyzed a specific circuit in a version of OpenAIs GPT-2, an older large language model that OpenAI released in 2019. But Anthropic has now analyzed a number of different circuits as a far larger and far more complex model carries out multiple tasks. Anthropic is very capable at applying scale to a problem, says Merullo.Eden Biran, who studies large language models at Tel Aviv University, agrees. Finding circuits in a large state-of-the-art model such as Claude is a nontrivial engineering feat, he says. And it shows that circuits scale up and might be a good way forward for interpreting language models.Circuits chain together different partsor componentsof a model. Last year, Anthropic identified certain components inside Claude that correspond to real-world concepts. Some were specific, such as Michael Jordan or greenness; others were more vague, such as conflict between individuals. One component appeared to represent the Golden Gate Bridge. Anthropic researchers found that if they turned up the dial on this component, Claude could be made to self-identify not as a large language model but as the physical bridge itself.The latest work builds on that research and the work of others, including Google DeepMind, to reveal some of the connections between individual components. Chains of components are the pathways between the words put into Claude and the words that come out.Its tip-of-the-iceberg stuff. Maybe were looking at a few percent of whats going on, says Batson. But thats already enough to see incredible structure.Growing LLMsResearchers at Anthropic and elsewhere are studying large language models as if they were natural phenomena rather than human-built software. Thats because the models are trained, not programmed.They almost grow organically, says Batson. They start out totally random. Then you train them on all this data and they go from producing gibberish to being able to speak different languages and write software and fold proteins. There are insane things that these models learn to do, but we dont know how that happened because we didnt go in there and set the knobs.Sure, its all math. But its not math that we can follow. Open up a large language model and all you will see is billions of numbersthe parameters, says Batson. Its not illuminating.Anthropic says it was inspired by brain-scan techniques used in neuroscience to build what the firm describes as a kind of microscope that can be pointed at different parts of a model while it runs. The technique highlights components that are active at different times. Researchers can then zoom in on different components and record when they are and are not active.Take the component that corresponds to the Golden Gate Bridge. It turns on when Claude is shown text that names or describes the bridge or even text related to the bridge, such as San Francisco or Alcatraz. Its off otherwise.Yet another component might correspond to the idea of smallness: We look through tens of millions of texts and see its on for the word small, its on for the word tiny, its on for the word petite, its on for words related to smallness, things that are itty-bitty, like thimblesyou know, just small stuff, says Batson.Having identified individual components, Anthropic then follows the trail inside the model as different components get chained together. The researchers start at the end, with the component or components that led to the final response Claude gives to a query. Batson and his team then trace that chain backwards.Odd behaviorSo: What did they find? Anthropic looked at 10 different behaviors in Claude. One involved the use of different languages. Does Claude have a part that speaks French and another part that speaks Chinese, and so on?The team found that Claude used components independent of any language to answer a question or solve a problem and then picked a specific language when it replied. Ask it What is the opposite of small? in English, French, and Chinese and Claude will first use the language-neutral components related to smallness and opposites to come up with an answer. Only then will it pick a specific language in which to reply. This suggests that large language models can learn things in one language and apply them in other languages.Anthropic also looked at how Claude solved simple math problems. The team found that the model seems to have developed its own internal strategies that are unlike those it will have seen in its training data. Ask Claude to add 36 and 59 and the model will go through a series of odd steps, including first adding a selection of approximate values (add 40ish and 60ish, add 57ish and 36ish). Towards the end of its process, it comes up with the value 92ish. Meanwhile, another sequence of steps focuses on the last digits, 6 and 9, and determines that the answer must end in a 5. Putting that together with 92ish gives the correct answer of 95.And yet if you then ask Claude how it worked that out, it will say something like: I added the ones (6+9=15), carried the 1, then added the 10s (3+5+1=9), resulting in 95. In other words, it gives you a common approach found everywhere online rather than what it actually did. Yep! LLMs are weird. (And not to be trusted.)The steps that Claude 3.5 Haiku used to solve a simple math problem were not what Anthropic expectedtheyre not the steps Claude claimed it took either. ANTHROPICThis is clear evidence that large language models will give reasons for what they do that do not necessarily reflect what they actually did. But this is true for people too, says Batson: You ask somebody, Why did you do that? And theyre like, Um, I guess its because I was . You know, maybe not. Maybe they were just hungry and thats why they did it.Biran thinks this finding is especially interesting. Many researchers study the behavior of large language models by asking them to explain their actions. But that might be a risky approach, he says: As models continue getting stronger, they must be equipped with better guardrails. I believeand this work also showsthat relying only on model outputs is not enough.A third task that Anthropic studied was writing poems. The researchers wanted to know if the model really did just wing it, predicting one word at a time. Instead they found that Claude somehow looked ahead, picking the word at the end of the next line several words in advance.For example, when Claude was given the prompt A rhyming couplet: He saw a carrot and had to grab it, the model responded, His hunger was like a starving rabbit. But using their microscope, they saw that Claude had already hit upon the word rabbit when it was processing grab it. It then seemed to write the next line with that ending already in place.This might sound like a tiny detail. But it goes against the common assumption that large language models always work by picking one word at a time in sequence. The planning thing in poems blew me away, says Batson. Instead of at the very last minute trying to make the rhyme make sense, it knows where its going.I thought that was cool, says Merullo. One of the joys of working in the field is moments like that. Theres been maybe small bits of evidence pointing toward the ability of models to plan ahead, but its been a big open question to what extent they do.Anthropic then confirmed its observation by turning off the placeholder component for rabbitness. Claude responded with His hunger was a powerful habit. And when the team replaced rabbitness with greenness, Claude responded with freeing it from the gardens green.Anthropic also explored why Claude sometimes made stuff up, a phenomenon known as hallucination. Hallucination is the most natural thing in the world for these models, given how theyre just trained to give possible completions, says Batson. The real question is, How in Gods name could you ever make it not do that?The latest generation of large language models, like Claude 3.5 and Gemini and GPT-4o, hallucinate far less than previous versions, thanks to extensive post-training (the steps that take an LLM trained on the internet and turn it into a usable chatbot). But Batsons team was surprised to find that this post-training seems to have made Claude refuse to speculate as a default behavior. When it did respond with false information, it was because some other component had overridden the dont speculate component.This seemed to happen most often when the speculation involved a celebrity or other well-known entity. Its as if the amount of information available pushed the speculation through, despite the default setting. When Anthropic overrode the dont speculate component to test this, Claude produced lots of false statements about individuals, including claiming that Batson was famous for inventing the Batson principle (he isnt).Still unclearBecause we know so little about large language models, any new insight is a big step forward. A deep understanding of how these models work under the hood would allow us to design and train models that are much better and stronger, says Biran.But Batson notes there are still serious limitations. Its a misconception that weve found all the components of the model or, like, a Gods-eye view, he says. Some things are in focus, but other things are still uncleara distortion of the microscope.And it takes several hours for a human researcher to trace the responses to even very short prompts. Whats more, these models can do a remarkable number of different things, and Anthropic has so far looked at only 10 of them.Batson also says there are big questions that this approach wont answer. Circuit tracing can be used to peer at the structures inside a large language model, but it wont tell you how or why those structures formed during training. Thats a profound question that we dont address at all in this work, he says.But Batson sees this as the start of a new era in which it is possible, at last, to find real evidence for how these models work: We dont have to be, like: Are they thinking? Are they reasoning? Are they dreaming? Are they memorizing? Those are all analogies. But if we can literally see step by step what a model is doing, maybe now we dont need analogies.

Seth Rogen has a new show about the business of Hollywood called "The Studio," and it may be Apple TV+'s next "Severance," with the show receiving stratospheric reviews.Seth Rogen in 'The Studio' - Image Credit: Apple TV+The first two episodes of "The Studio" which skewers the behind-the-scenes nature of working in Hollywood, has received almost universal acclaim from critics, professional and amateur alike. As of March 27, Rotten Tomatoes reports the show at 96% "Certified Fresh," with 22 out of 23 "Top Critics" reviews considering it "Fresh" and only one listed as "Rotten" on the "Tomatometer."While critics are mostly raving about it, audiences take a slightly more muted approach. The site's "Popcornmeter," which uses from reviews from audiences, gives it an average score of 3.7 out of 5 stars, with 69% of reviews rating it at 3.5 stars or above. Continue Reading on AppleInsider | Discuss on our Forums

Apple has seeded a second release candidate for macOS Sequoia 15.4, as the updates to the company's operating systems get closer to a public release.MacBook Air running macOSApple initially introduced the first round of release candidates for macOS 15.4, iOS 18.4, iPadOS 18.4, tvOS 18.4, watchOS 11.4, and visionOS 2.4 on March 24. Three days later, it has introduced a second RC build of macOS 15.4 to test out.It's not uncommon for there to be multiple release candidate builds of software before it is released to the public. While a release candidate is supposed to have the same content as the final release, sometimes there is a missing or incorrect element that needs addressing urgently. Continue Reading on AppleInsider | Discuss on our Forums

As the important task of debris removals continues in the devastated areas of Altadena, the LA Times has picked up on the mission to recover historic tiles from the surviving mantelpieces of burned homes. There, a patchwork battalion of masons, homeowners, and other volunteers are working diligently under the monicker Save the Tiles to save the sometimes hundred-year-old vestiges of the early 20th century Arts and Crafts movement and Los Angeles architectural past from potential looters.The community at the foot of theSan Gabriel Mountains lost up to 9,400 homes during Januarys Los Angeles Fires. Save the Tiles' effort is taking place side-by-side with aseparate SoCalNOMA effortseparate SoCalNOMA effortwhose intention is to preserve Altadenas historic Black communitys stake in the face of developer encroachment.

As the new Trump Administration passes 60 days in office, tariffscontinue to form a cornerstone of the Administrations trade policy. On March 12th, the US imposed steel and aluminum tariffs on Canada with more potentially coming on April 2nd. Meanwhile, a trade probe on US lumber imports being carried out by the US Commerce Secretary may yet see tariffs imposed on lumber products from countries including Canada.Archinect has reported extensively on the impact of such tariffs, whether implemented or threatened, on the US architecture and construction sector. Construction input prices have risen, with the ABC noting in February that tariff threats caused a rapid escalation in material prices. The ABC has also previously warned that a trade war with Canada and Mexico could impact construction spending, while the AIA said last week that a continuing decline in their Architecture Billings Index is being partly fuelled by tariff uncertainty.The developments present two urgent situa...