Maths can be a scary topic for people. Many of you want to work in machine learning, but the maths skills needed may seem overwhelming. I am here to tell you that it’s nowhere as intimidating as you may think and to give you a roadmap, resources, and advice on how to learn math effectively. Let’s get into it! Do you need maths for machine learning? I often get asked: Do you need to know maths to work in machine learning? The short answer is generally yes, but the depth and extent of maths you need to know depends on the type of role you are going for. A research-based role like: Research Engineer — Engineer who runs experiments based on research ideas. Research Scientist — A full-time researcher on cutting edge models. Applied Research Scientist — Somewhere between research and industry. You will particularly need strong maths skills. It also depends on what company you work for. If you are a machine learning engineer or data scientist or any tech role at: Deepmind Microsoft AI Meta Research Google Research You will also need strong maths skills because you are working in a research lab, akin to a university or college research lab. In fact, most machine learning and AI research is done at large corporations rather than universities due to the financial costs of running models on massive data, which can be millions of pounds. For these roles and positions I have mentioned, your maths skills will need to be a minimum of a bachelor’s degree in a subject such as math, physics, computer science, statistics, or engineering. However, ideally, you will have a master’s or PhD in one of those subjects, as these degrees teach the research skills needed for these research-based roles or companies. This may sound heartening to some of you, but this is just the truth from the statistics. According to a notebook from the 2021 Kaggle Machine Learning & Data Science Survey, the research scientist role is highly popular among PhD and doctorates. Source. And in general, the higher your education the more money you will earn, which will correlate with maths knowledge. Source. However, if you want to work in the industry on production projects, the math skills needed are considerably less. Many people I know working as machine learning engineers and data scientists don’t have a “target” background. This is because industry is not so “research” intensive. It’s often about determining the optimal business strategy or decision and then implementing that into a machine-learning model. Sometimes, a simple decision engine is only required, and machine learning would be overkill. High school maths knowledge is usually sufficient for these roles. Still, you may need to brush up on key areas, particularly for interviews or specific specialisms like reinforcement learning or time series, which are quite maths-intensive. To be honest, the majority of roles are in industry, so the maths skills needed for most people will not be at the PhD or master’s level.  But I would be lying if I said these qualifications do not give you an advantage. What maths do you need to know? There are three core areas you need to know: Statistics Calculus Linear Algebra Statistics I may be slightly biased, but statistics is the most important area you should know and put the most effort into understanding. Most machine learning originated from statistical learning theory, so learning statistics will mean you will inherently learn machine learning or its basics. These are the areas you should study: Descriptive Statistics — This is useful for general analysis and diagnosing your models. This is all about summarising and portraying your data in the best way. Averages: Mean, Median, Mode Spread: Standard Deviation, Variance, Covariance Plots: Bar, Line, Pie, Histograms, Error Bars Probability Distributions — This is the heart of statistics as it defines the shape of the probability of events. There are many, and I mean many, distributions, but you certainly don’t need to learn all of them. Normal Binomial Gamma Log-normal Poisson Geometric Probability Theory — As I said earlier, machine learning is based on statistical learning, which comes from understanding how probability works. The most important concepts are Maximum likelihood estimation Central limit theorem Bayesian statistics Hypothesis Testing —Most real-world use cases of data and machine learning revolve around testing. You will test your models in production or carry out an A/B test for your customers; therefore, understanding how to run hypothesis tests is very important. Significance Level Z-Test T-Test Chi-Square Test Sampling Modelling & Inference —Models like linear regression, logistic regression, polynomial regression, and any regression algorithm originally came from statistics, not machine learning. Linear Regression Logistic Regression Polynomial Regression Model Residuals Model Uncertainty Generalised Linear Models Calculus Most machine learning algorithms learn from gradient descent in one way or another. And, gradient descent has its roots in calculus. There are two main areas in calculus you should cover: Differentiation What is a derivative? Derivatives of common functions. Turning point, maxima, minima and saddle points. Partial derivatives and multivariable calculus. Chain and product rules. Convex vs non-convex differentiable functions. Integration What is integration? Integration by parts and substitution. The integral of common functions. Integration of areas and volumes. Linear Algebra Linear algebra is used everywhere in machine learning, and a lot in deep learning. Most models represent data and features as matrices and vectors. Vectors  What are vectors Magnitude, direction Dot product Vector product Vector operations (addition, subtraction, etc) Matrices  What is a matrix Trace Inverse Transpose Determinants Dot product Matrix decomposition Eigenvalues & Eigenvectors  Finding eigenvectors Eigenvalue decomposition Spectrum analysis Best Resources There are loads of resources, and it really comes down to your learning style. If you are after textbooks, then you can’t go wrong with the following and is pretty much all you need: Practical Statistics For Data Scientist — I recommend this book all the time and for good reason. This is the only textbook you realistically need to learn the statistics for Data Science and machine learning. Mathematics for Machine Learning — As the name implies, this textbook will teach the maths for machine learning. A lot of the information in this book may be overkill, but your maths skills will be excellent if you study everything. If you want some online courses, I have heard good things about the following ones. Mathematics for Machine Learning and Data Science Specialisation — This course is by DeepLearning.AI, the same people who made the Machine Learning Specialisation, arguably the best machine learning course. Learning Advice The amount of maths content you need to learn may seem overwhelming, but don’t worry. The main thing is to break it down step by step. Pick one of the three: statistics, Linear Algebra or calculus. Look at the things I wrote above you need to know and choose one resource. It doesn’t have to be any of the ones I recommended above. That’s the initial work done. Don’t overcomplicate by looking for the “best resource” because such a thing doesn’t exist. Now, start working through the resources, but don’t just blindly read or watch the videos. Actively take notes and document your understanding. I personally write blog posts, which essentially employ the Feynman technique, as I am, in a way, “teaching” others what I know. Writing blogs may be too much for some people, so just make sure you have good notes, either physically or digitally, that are in your own words and that you can reference later. The learning process is generally quite simple, and there have been studies done on how to do it effectively. The general gist is: Do a little bit every day Review old concepts frequently (spaced repetition) Document your learning It’s all about the process; follow it, and you will learn! Another Thing! Join my free newsletter, Dishing the Data, where I share weekly tips, insights, and advice from my experience as a practicing Machine Learning engineer. Plus, as a subscriber, you’ll get my FREE Data Science / Machine Learning Resume Template! Dishing The Data | Egor Howell | SubstackAdvice and learnings on data science, tech and entrepreneurship. Click to read Dishing The Data, by Egor Howell, a…newsletter.egorhowell.com Connect with me YouTube, LinkedIn, Instagram Book a 1:1 mentoring call The post How to Learn the Math Needed for Machine Learning appeared first on Towards Data Science.

News Marathon Uses Decals and Visual Style Stolen From Independent Artist, Confirms Bungie Bungie said that it will be implementing stricter checks to ensure that all artwork used in its assets are attributed correctly. Posted By Joelle Daniels | On 16th, May. 2025 As Bungie gets ready for the September 23 release of its PvPvE extraction shooter Marathon, the studio has found itself embroiled in a new controversy. A visual artist, going by Antireal, has taken to social media platform X accusing Bungie of stealing the artistic style and design elements from projects she had worked on back in 2017. In her post, Antireal said, “the Marathon alpha released recently and its environments are covered with assets lifted from poster designs I made in 2017.” She went on to say that, while Bungie wasn’t obligated to offer her a job for her design work, the company outright stealing her ideas without appropriate attribution or compensation wasn’t a good thing. “Bungie is of course not obligated to hire me when making a game that draws overwhelmingly from the same design language I have refined for the last decade, but clearly my work was good enough to pillage for ideas and plaster all over their game without pay or attribution,” wrote Antireal. “I don’t have the resources nor the energy to spare to pursue this legally but I have lost count of the number of times a major company has deemed it easier to pay a designer to imitate or steal my work than to write me an email.” “In 10 years I have never made a consistent income from this work and I am tired of designers from huge companies moodboarding and parasitising my designs while I struggle to make a living.” Antireal’s posts are accompanied by a host of images showcasing some of her work, which does bear a striking resemblance to the overall design language we’ve seen from the various trailers and artwork released for Marathon over the last couple of years. Some elements from Antireal’s art seem to be outright lifted and pasted into Marathon, including words like “Aleph” which feature prominently in the artist’s work. In response, Bungie took to X to confirm that Antireal’s work was indeed seemingly used during the development of Marathon. However, the studio notes that this issue was unknown for its art team, and that it is taking investigating how this kind of slip up could happen. Bungie has referred to the use of Antireal’s work in Marathon’s textures an “unauthorised use of artist decals.” “This issue was unknown by our existing art team, and we are still reviewing how this oversight occurred,” wrote the studio. “We take matters like this very seriously. We have reached out to [Antireal] to discuss this issue and are committed to do right by the artist.” “As a matter of policy, we do not use the work of artists without their permission.” Bungie has also said that, moving forward, it will be conducting a thorough review of its in-game assets to ensure that there are stricter checks to document artist contributions. “We value the creativity and dedication of all artists who contribute to our games, and we are committed to doing right by them,” it wrote. “Thank you for bringing this to our attention.” Marathon is coming to PC, PS5 and Xbox Series X/S, and is currently undergoing playtests. Tagged With: Elden Ring: Nightreign Publisher:Bandai Namco Developer:FromSoftware Platforms:PS5, Xbox Series X, PS4, Xbox One, PCView More DOOM: The Dark Ages Publisher:Bethesda Softworks Developer:id Software Platforms:Xbox Series X, PS5, PCView More Death Stranding 2: On the Beach Publisher:Sony Developer:Kojima Productions Platforms:PS5View More Amazing Articles You Might Want To Check Out! Trails in the Sky 1st Chapter Trailer Showcases Opening Cutscene Remade in 3D Watch the first meeting between Joshua and Estelle with full animations and facial expressions before the rema... Tainted Grail: The Fall of Avalon – Everything You Need to Know The new indie 'scrolls-like' is now feature complete and looks stunning. Grand Theft Auto 6 Development Began “In Earnest” in 2020 – Take-Two Interactive The statement indicates that work had already begun on GTA 6 prior to 2020, which would back a former dev sayi... Mario Kart World – Everything You Need to Know [drop-cap]A[/drop-cap]fter what felt like an eternity of waiting, the next Mario Kart is finally almost upon... Read More Marathon Uses Decals and Visual Style Stolen From Independent Artist, Confirms Bungie Bungie said that it will be implementing stricter checks to ensure that all artwork used in its assets are att... Avowed to Receive New Game Plus, Photo Mode, and New Weapons This Year The next update due this Summer includes additional NPC abilities and behaviors, new Unique weapons and armor,... View More

pixel_dreams - Fotolia News Security tests reveal serious vulnerability in government’s One Login digital ID system A ‘red teaming’ exercise to simulate cyber attacks on the government’s flagship digital identity system has found that One Login can be compromised without detection By Bryan Glick, Editor in chief Published: 16 May 2025 12:37 External security tests on the government’s flagship digital identity system, Gov.uk One Login, have found serious vulnerabilities in the live service, Computer Weekly has learned. A “red teaming” exercise conducted in March by IT security consultancy Cyberis discovered that privileged access to One Login can be compromised without detection by security monitoring tools. According to Cyberis, red teaming tests the resilience of systems by simulating the tactics, techniques and procedures of cyber attackers to show how well an organisation can detect and respond to an incident. Computer Weekly has been asked by the Department for Science, Innovation and Technology (DSIT) not to reveal further details of the vulnerability while the Government Digital Service (GDS) seeks to fix the problem. Compromising the highest levels of access to a system risks exposing personal data and software code to any cyber attackers able to exploit the vulnerability. A government spokesperson said: “Delivering best practice, we routinely conduct red teaming exercises to test security infrastructure. Where issues are found, we work urgently to resolve them.” The existence of a serious current vulnerability will raise further concerns over the security of One Login, which is intended to be the way that citizens prove their identity and log in to most online government services. There are already six million users of the system, and it is used to access more than 50 online services. Last month, Computer Weekly revealed that GDS was warned by the Cabinet Office in November 2022 and the National Cyber Security Centre (NCSC) in September 2023, that One Login had “serious data protection failings” and “significant shortcomings” in information security that could increase the risk of data breaches and identity theft. GDS said the concerns were “outdated” and arose “when the technology was in its infancy in 2023”, despite One Login being used at that time to support live services. “We have worked to address all these concerns as evidenced by multiple external independent assessments. Any suggestion otherwise is unfounded,” said a spokesperson, at the time. A whistleblower first raised security concerns about One Login within GDS as long ago as July 2022. The issues identified included system administration being performed through non-compliant devices with a risk of transmitting security vulnerabilities, such as malware or phishing attacks, that could compromise the live system. The NCSC recommends that system administration for key government services should be conducted from a dedicated device used only for that purpose, known as a privileged access workstation (PAW), or alternatively to use only “browse down” devices, where the security level of the device is always the same or greater than the system being managed. The whistleblower warned that a lack of PAWs and use of browse-up administration were significant risks. Computer Weekly subsequently revealed that the One Login team has yet to fully meet NCSC guidelines – the system only complies with 21 of the 39 outcomes detailed in the NCSC Cyber Assessment Framework (CAF) – an improvement on the five outcomes it successfully followed a year ago. The One Login development team is also yet to fully implement the government’s Secure by Design practices, although GDS said the system “meets these principles”. Earlier this week, we further revealed that One Login had lost its certification against the government’s own trust framework for digital identity systems, after a key technology supplier allowed its certification to lapse and, as a result, One Login was removed from the official accreditation scheme. In a meeting with private sector digital identity providers this week (Wednesday 14 May), DSIT secretary of state Peter Kyle explained how One Login will underpin the forthcoming Gov.uk Wallet, which will be used to deliver digital versions of key government documents, such as driving licences. Kyle talked about the “rapid journey” he hopes the government will take in delivering digital identity services for citizens and stressed the importance that such systems are “delivered safely [and] securely”. The government spokesperson added: “Gov.uk One Login follows the highest security standards for government and private sector services – including dedicated 24/7 eyes-on monitoring and incident response. As the public rightly expects, protecting the security of government services and the data and privacy of users to keep pace with the changing cyber threat landscape is paramount.” Questions are also being asked in Parliament about the security of One Login. In recent weeks, Liberal Democrat peer and digital spokesman Tim Clement-Jones and Conservative peer Simone Finn have separately submitted Parliamentary questions to DSIT asking for reassurances about the system. Finn asked whether the government has “quantified the likelihood and potential impact of insider threats, unauthorised privileged access, and production environment compromise within One Login”. In response, DSIT minister for the future digital economy and online safety, peer Maggie Jones, said: “The Gov.uk One Login team collaborates closely with the NCSC to assess and mitigate risks associated with insider threats, unauthorised privileged access, and production environment compromise, aligning with the Cyber Assessment Framework outlined in the Government Cyber Security Strategy 2022-2030. “While assessments of insider threats have been made, copies of these assessments will not be placed in the Library of the House, as they are part of ongoing security measures and internal governance processes.” Clement-Jones asked: “What steps [the government is] taking to address security issues in the One Login digital identification system?” Jones replied: “One Login follows the highest security standards for government and private sector services. As the public rightly expects, protecting the security of government services and the data and privacy of users to keep pace with the changing cyber threat landscape is paramount. “Security best practice is followed with a number of layered security controls which include: Security clearances for staff with ‘Security Check’ clearance required for all developers with production access; identity and access management controls that block staff from viewing or altering personal information; a secure by design and compartmentalised system architecture; technical controls around building and deployments; logging and monitoring to alert on access to environments that contain personally identifiable information; and robust procedures for addressing any unauthorised or unaccounted for access.” Speaking to Computer Weekly about the security concerns, Clement-Jones said: “How is the government’s flagship digital identity system failing to meet standards so badly, given that it is expected to shortly form an essential part of our immigration controls? We need answers and quickly.”  Read more about One Login Companies House goes live with One Login ID verification – People can verify their identity with Companies House using Gov.uk One Login as the central government body becomes the 36th service to start using the digital identity system. GDS goes serverless to bring personalisation to online government services with One Login – GDS has opened up about the reasons why it’s opted for a serverless infrastructure to underpin One Login, and how it hopes the system will provide UK citizens with a more personalised experience. One Login digital identity project makes headway – Government services are lining up to work with the GDS on its One Login digital identity system, according to its director of digital identity, Natalie Jones. In The Current Issue: UK MoJ crime prediction algorithms raise serious concerns Interview: Markus Schümmelfeder, CIO, Boehringer Ingelheim Download Current Issue GraphQL as an ‘essential protocol’ for AI-API orchestration – CW Developer Network Mind the insight-to-impact gap, Qlik captures analytics ‘in the moment’ – CW Developer Network View All Blogs

As Bungie gets ready for the September 23 release of its PvPvE extraction shooter Marathon, the studio has found itself embroiled in a new controversy. A visual artist, going by Antireal, has taken to social media platform X accusing Bungie of stealing the artistic style and design elements from projects she had worked on back in 2017. In her post, Antireal said, “the Marathon alpha released recently and its environments are covered with assets lifted from poster designs I made in 2017.” She went on to say that, while Bungie wasn’t obligated to offer her a job for her design work, the company outright stealing her ideas without appropriate attribution or compensation wasn’t a good thing. “Bungie is of course not obligated to hire me when making a game that draws overwhelmingly from the same design language I have refined for the last decade, but clearly my work was good enough to pillage for ideas and plaster all over their game without pay or attribution,” wrote Antireal. “I don’t have the resources nor the energy to spare to pursue this legally but I have lost count of the number of times a major company has deemed it easier to pay a designer to imitate or steal my work than to write me an email.” “In 10 years I have never made a consistent income from this work and I am tired of designers from huge companies moodboarding and parasitising my designs while I struggle to make a living.” Antireal’s posts are accompanied by a host of images showcasing some of her work, which does bear a striking resemblance to the overall design language we’ve seen from the various trailers and artwork released for Marathon over the last couple of years. Some elements from Antireal’s art seem to be outright lifted and pasted into Marathon, including words like “Aleph” which feature prominently in the artist’s work. In response, Bungie took to X to confirm that Antireal’s work was indeed seemingly used during the development of Marathon. However, the studio notes that this issue was unknown for its art team, and that it is taking investigating how this kind of slip up could happen. Bungie has referred to the use of Antireal’s work in Marathon’s textures an “unauthorised use of artist decals.” “This issue was unknown by our existing art team, and we are still reviewing how this oversight occurred,” wrote the studio. “We take matters like this very seriously. We have reached out to [Antireal] to discuss this issue and are committed to do right by the artist.” “As a matter of policy, we do not use the work of artists without their permission.” Bungie has also said that, moving forward, it will be conducting a thorough review of its in-game assets to ensure that there are stricter checks to document artist contributions. “We value the creativity and dedication of all artists who contribute to our games, and we are committed to doing right by them,” it wrote. “Thank you for bringing this to our attention.” Marathon is coming to PC, PS5 and Xbox Series X/S, and is currently undergoing playtests. bungie is of course not obligated to hire me when making a game that draws overwhelmingly from the same design language i have refined for the last decade, but clearly my work was good enough to pillage for ideas and plaster all over their game without pay or attribution. pic.twitter.com/G3FbPtbPJD— N² (@4nt1r34l) May 15, 2025 This issue was unknown by our existing art team, and we are still reviewing how this oversight occurred.We take matters like this very seriously. We have reached out to @4nt1r34l to discuss this issue and are committed to do right by the artist.— MarathonDevTeam (@MarathonDevTeam) May 16, 2025 To prevent similar issues in the future, we are conducting a thorough review of our in-game assets, specifically those done by the former Bungie artist, and implementing stricter checks to document all artist contributions.— MarathonDevTeam (@MarathonDevTeam) May 16, 2025

Macworld At a glanceExpert's Rating Pros Outstanding performance Good display Useful battery life Bright display with anti-reflective coating Excellent build quality Cons High price for 5G No OLED No camera flash Few innovations since 2022 Our Verdict While we’d have liked to see the inclusion of an M4 chip instead of the M3, Apple’s new iPad Air is well worth a recommendation, offering an excellent display and strong performance for years to come. Paired with a good keyboard case, it could even replace your MacBook. Our only caveat is that the previous model with the M2 chip, which you can find at a lower price, might be an even better buy. Price When Reviewed This value will show the geolocated pricing text for product undefined Best Pricing Today Best Prices Today: iPad Air 11 Zoll (M3, 2025) New Refurbished Retailer Price €648 View Deal €669 View Deal €669 View Deal €673 View Deal €630.2 View Deal 629,00 € View Deal 634,60 € View Deal 638,90 € View Deal 644,00 € View Deal 648,00 € View Deal 648,00 € View Deal 660,05 € View Deal 679,99 € View Deal €680.38 View Deal 699,00 € View Deal Price comparison from over 24,000 stores worldwide View more prices Product Price iPad Air M3 11 (2025) 7. Generation 128 GB – WLAN – Blau €649 View Deal Price comparison from Backmarket The iPad Air has consistently been the best choice for most users, and this remains true of the new version, which offers the best balance of price and features among Apple’s tablets. Thanks to its new M3 chip, it has become significantly more powerful than the previous M2 model. However, this is otherwise quite a minor update. Surely you would expect a new iPad to boast the very latest chip (the M4 is currently available on the iPad Pro), a new design and some cool new features? Regardless of expectations, the iPad Air has remained visually unchanged for three years; this one looks exactly the same as the models from last year and 2022. Apple hasn’t even given the Air any new colors for 2025. The iPad Pro, by contrast, was given a more innovative, thinner design last year, higher performance and an impressive OLED display. The Pro models bring in the most money, which might suggest deliberate neglect of the Air models. Still, the iPad Air could be back next year with a new design. For now, let’s see if the new M3 Air is the right tablet for you. Design and build quality: Superb The iPad Air is an elegant device with first-class workmanship. The tried-and-tested aluminum frame makes the tablet solid and attractive, and iPads can generally be relied upon to remain presentable even after several years of intensive use. Externally, you won’t notice any difference from the previous model. Even the dimensions are exactly the same, down to a tenth of a millimeter; only the weight of the 11-inch model has fallen by 2g. The color options are again Space Gray, Starlight, blue, and purple. The colors are significantly less gaudy than those of the 11th-gen iPad and struck us as very subtle: even blue and purple look “grown-up” and the Starlight finish is barely distinguishable from silver. The 11-inch model in particular feels light and sturdy at the same time. With a weight of 460g, this can be used with one hand, while the 13-inch model is significantly heavier at 616g. However, even the large model is just half the weight of a MacBook Air. The Air has an appealing and successful design.Eugen Wegmann Benefits of the M3 chip The most important new feature of the iPad Air is the M3 chip, but you shouldn’t expect too much. On its website, Apple promises a significant increase in performance. The new Air, it says, is “nearly 2x faster than the iPad Air with M1 chip.” That’s not wrong, but it’s also a little misleading; the M1 model came out back in spring 2022. The comparison with the M2 model from last year, by contrast, is less flattering: in our tests the performance leap was rarely noticeable in everyday use. So we should think of this as a small refresh for the iPad Air, a kind of “speed bump.” For a proper speed boost, Apple would have needed to install the M4, and it’s pretty obvious why the company didn’t. The performance gap between the Air and the expensive top model has to be maintained. Still, while it doesn’t represent a startling improvement on the M2, don’t entirely write off the benefits of the new chip. Not only does it have a higher clock rate, it also offers new features such as support for hardware ray tracing–an important consideration for certain high-end games. The new media engine of the M3 supports higher resolutions and can now play HEVC videos up to 8K and H264 videos up to 4K, as well as using the ProRes and ProRES RAW formats. AV1 decoding is also supported, which could be very useful if you want to edit videos on the iPad. Performance: Moderate improvements The iPad doesn’t feature the same M3 chip that Apple used in the MacBook Air, and which was available in two versions (8-core CPU and 8-core GPU, and 8-core CPU and 10-core GPU). Instead, the iPad Air gets a third version of the M3 with an 8-core CPU and a 9-core GPU. Graphics performance therefore sits between the two earlier versions. The M3 brings a solid performance leap in single-core performance, since the M3 chip is clocked significantly higher than the M2. In the Geekbench 6 benchmark the M3 iPad scored 3,019 points to the M2 iPad’s 2,584 points. That’s almost 15 per cent more performance, which is available for tasks such as web surfing or Apple Intelligence. In the multi-core test, the M3 scored 11,749 points to the M2’s 10,074, an improvement of around 17 percent. This higher multiprocessor performance is helpful for more complex apps, such as image and video editing. The Geekbench 5 benchmark illustrates the current models’ improved performance.Foundry But when it comes to GPU performance the new chip offers far less improvement. In the 3D Mark Wildlife Extreme test, the M2 scored 6,443 points and the M3 managed just three points more at 6,446. In the Geekbench GPU test, things were a little better, at 46,007 points vs 41,902, but still below a 10 percent increase. However, the performance remains excellent in comparison with other iPads. The iPad (A16), which was launched at the same time, is less than half as quick in terms of graphics performance and scored just 20,123 points in Geekbench’s graphics benchmark. We also test how long a CPU can sustain top performance. (Performance levels often drop rapidly when the device heats up, as iPads have to cope without a fan.) According to our APSI test running 20 minutes of full load, the Air’s throttling remains within acceptable limits at 86.57 percent. Note that the new iPad Pro is better at 90.62 percent, and the 11th-gen iPad, with its far less powerful A16 CPU, heats up even less and performs even better with 93.94 percent. The Air supports the fast new Wi-Fi 6E standard, whereas the iPad A16 still has to make do with Wi-Fi 6. If you pay extra for cellular connectivity, which adds $150 to the price, then you can enjoy 5G support via eSIM. Storage performance: Nowhere near the iPad Pro The M3 gets the headlines, but don’t underestimate the importance of the iPad’s internal SSD when it comes to performance. Our 1TB model in the test delivered what we’d regard as adequate performance, achieving 1,646 MB/s when reading and 1,380 MB/s when writing. But this doesn’t come close to the performance of an iPad Pro M4, where we measured 3,362 MB/s and 2,017 MB/s for reading and writing respectively. It’s also possible to connect an external SSD via USB-C. As an experiment, we transferred some large video files between the iPad Air and a Lacie Thunderbolt SSD. We saw speeds of up to 700 MB/s via the Files app, which puts the Air somewhere in the iPad midfield: the 11th-gen iPad managed only 37 MB/s in the same test, but the iPad Pro M4 achieved 2,200 MB/s. Display: Lacking some Pro features As usual, the biggest difference between the iPad, iPad Air and iPad Pro product lines is the screen. With the iPad Air, you get a good 11- or 13-inch panel with IPS technology, which is still common among tablets and laptops. The resolution is 2360 x 1640 pixels (11 inches) or 2732 x 2048 (13 inches). At 600 nits, the 13-inch model is brighter than the 11-inch model, which only reaches a maximum of 500 nits. However, both displays are bright enough to work with outdoors. Compared to the cheaper 11th-gen iPad, the display scores points with a useful anti-glare coating, which means it reflects significantly less than the basic iPad. More advanced technologies such as OLED (and previously mini-LED) are unfortunately reserved for the Pro models, which support HDR content better as a result. An OLED display can also be considerably more frugal in terms of battery consumption. In addition to the sRGB color standard, the iPad Air supports the larger P3 color space, which is primarily used in the video sector. A matte display is only available as an option for the iPad Pro, and ProMotion is a Pro exclusive too. Gamers would certainly appreciate a faster display, but will have to make do with 60Hz on the Air. For most users, however, the lack of ProMotion is not a big deal. Size: Should you get the 11-inch or 13-inch model? For a long time, if you wanted a big 13-inch screen, you had to go for the iPad Pro. But since last year, the Air has also been available with a 13-inch screen too, and buyers are spoilt for choice. Ultimately, every buyer has to decide for themselves which form factor is better for them. Bigger doesn’t always mean better: the 11-inch model is easier to handle, especially if you’re using it on the move. It’s also significantly lighter. In my opinion, there are two main arguments in favour of the 13-inch Air. If you often use the device with an external keyboard, you’ll probably be more productive with the larger model. Managing several apps with Split View or Stage Manager is also more practical at this size. In my opinion, Stage Manager only really makes sense with these larger displays. On a 13-inch model, a single Safari window in full-screen mode is simply too large. With Stage Manager, you can have several apps open and work with Word, Mail, and Safari at the same time, just like on a Mac. In addition, an iPad with a 13-inch display is ideal as an external Mac display. If you use a small iMac as your main device, you can integrate the iPad as an additional display if required. While an 11-inch iPad is a little too small for this role, a 13-incher is really useful. My take is that the 11-inch model is better for consumption, and the 13-inch model is better for work. Battery life: Solid Battery life has hardly changed compared to the previous model. According to Coconut Battery, the battery of our (13-inch) test unit had a capacity of 9782 mAh, and this offers enough power for a day of office work and web surfing. Apple promises up to 10 hours of web browsing over Wi-Fi or video playback, but bear in mind that the company bases that figure on testing with a screen brightness of just 50 percent. In our tests we use maximum brightness, which is a lot more demanding. In our video test, where we let a current movie run in a continuous loop, the 13-inch Air lasted six and a half hours, which is a good result. In our second test, where a website is called up at rapid intervals, the Air lasted almost eight hours. This is all fine, but the unflattering comparison with the iPad Pro illustrates the drawbacks of an IPS screen: thanks to its efficient OLED display, the iPad Pro M4 lasted three hours longer in the surfing test and twice as long in the video test. The M3 chip can consume a lot of battery power when it’s at maximum load, for example when gaming or using 3D apps. This is a clear difference from iPads with a slow CPU such as the 10th-gen iPad. Having a slow CPU has plenty of drawbacks, but it reduces battery consumption. Keyboards, cases, and other accessories Apple has introduced a new keyboard alongside the new iPad Air, and this has led to a somewhat confusing range of accessories to choose from. Apple still sells a version of the Magic Keyboard for the iPad Air M2 (from $299 for the 11-inch version), in either black or white. But now there’s a new Magic Keyboard for the iPad Air M3, which is only available in white. The price for the 11-inch version is $269/£269, while the 13-inch version costs $319/£299. Apple has improved the keyboard compared to the previous version in a number of ways. For example, there’s a set of function keys, while the trackpad is larger. And it’s important to note that, rather confusingly and contrary to what its branding might suggest, this new accessory is compatible with the iPad Air M2 (and indeed the 4th- and 5th-gen Airs) as well as the iPad Air M3. However, there are a few compromises compared with the iPad Pro’s Magic Keyboard; only the Pro version offers a glass trackpad and backlit buttons. It’s rather annoying that if you ever switch from the iPad Air to the iPad Pro, you’ll also have to factor in the (not inconsiderable) cost of a new keyboard if you want to get the full benefit of the upgrade. This keyboard is missing a few features from the Pro Magic Keyboard.IDG There are no new Smart Folios, but Apple has dropped their prices. Most third-party cases for the previous iPad Air models will also be suitable for the new model, such as the Logitech Combo Touch. Finally, like its predecessor, the iPad supports both the USB-C Apple Pencil and the more advanced Pencil Pro. Older Pencils can no longer be used. External display: A handy option The M3 Air offers support for an external display with a resolution of up to 6K at 60Hz, compared to a maximum of 4K on the 11th-gen iPad. Don’t underestimate the value of this option. If you connect your iPad to an external display via USB-C with a mouse and keyboard connected, the Stage Manager display option is automatically activated. You can then also use the mouse and keyboard and more efficiently edit long texts or videos. Photo and video: Better at video chats than stills photography The iPad isn’t an obvious choice as a photographic companion, but the new Air, as usual, is equipped with cameras on the front and back. These don’t compare with current iPhone cameras, especially when taking photos indoors: images quickly become noisy. And unfortunately you have to do without a flash, which would have been useful when taking photos of documents. The front-facing camera is positioned on the long edge as has become standard practice for Apple, which is the ideal position for video conferencing. Plus the Center Stage feature allows you to choose between auto-zoom or a wide-angle image. The Air is well equipped for video chats. Video recordings with the main camera are still possible at up to 4K and 60fps, and the Air gets video image stabilisation. Finally, here’s one area where the Air outperforms the iPad Pro: its rear camera module barely protrudes, which is more practical in everyday use. Audio setup: Impressive quality I was impressed by the sound quality of the two speakers located on the sides, which remains strong even at higher volumes. This is not only useful for music and videos, but also for presentations and video calls. In most situations there’s no need to add external Bluetooth speakers. Two integrated microphones ensure good recording quality during video conferences. The iPad Pro offers even better sound quality thanks to its four speakers and five microphones, but the Air is already at a high level in this area. Summary: The sweet spot Overall, the iPad Air is convincing. The iPad Pro is even better, but far too expensive. The iPad A16, on the other hand, is significantly slower and has a poorer display. This feels like a sweet spot. For most users, in fact, the iPad Air is the ideal choice: fast, ergonomic, and equipped with a good display. Our only caveat is that there’s hardly any reason to buy this model rather than the cheaper previous model with an M2 chip, which is barely less powerful; what’s more, there are some excellent deals out there. The M2 Air is, therefore my personal recommendation. Whether you choose the 11- or 13-inch model is up to you. Tech specs iPad Air (M3, 2025)11 inches13 inchesPrice (base model, Wi-Fi)$599$799Price (base model, cellular)$749$949Display11-inch Liquid Retina13-inch Liquid RetinaResolution2360 x 1640, 264ppi2732 x 2048, 264ppiDisplay technologyLED, fully laminated, anti-reflective coatingLED, fully laminated, anti-reflective coatingColor spaceWide color gamut (P3), True ToneWide color gamut (P3), True ToneBrightness500 nits600 nitsChipApple M3Apple M3CPU8-core CPU: 4 performance and 4 efficiency8-core CPU: 4 performance and 4 efficiencyGraphics9-core GPU9-core GPUNeural Engine16-core16-coreMedia EngineHardware accelerated H.264 and HEVC, AV1 decodingHardware accelerated H.264 and HEVC, AV1 decodingMemory8GB8GBStorage capacity128GB/256GB/512GB/1TB128GB/256GB/512GB/1TBCameras12MP wide-angle, f/1.812MP wide angle, f/1.8Front camera12MP landscape format, Center Stage, f/2.012MP landscape format, Center Stage, f/2.0Speaker, microphoneStereo speakers in landscape format, 2 microphonesStereo speakers in landscape format, 2 microphonesSensorsTouch ID in the power buttonTouch ID in the power buttonConnectionsUSB 3 (up to 10 Gbit/s), Smart ConnectorUSB 3 (up to 10 Gbit/s), Smart ConnectorPencil (optional)Apple Pencil Pro, Apple Pencil (USB-C)Apple Pencil Pro, Apple Pencil (USB-C)Wi-FiWLAN 6E (802.11ax) with 2×2 MIMOWi-Fi 6E (802.11ax) with 2×2 MIMOBluetooth5.35.3Mobile data5G/LTE/UMTS/HSDPA5G/LTE/UMTS/HSDPASIM cardeSIMeSIMColorsSpace Gray, Starlight, purple, blueSpace Gray, Starlight, purple, blueBatteryLithium polymer, 28.93 WhLithium polymer, 36.59 WhSurfing Wi-Fi10 hours10 hoursMobile surfing9 hours9 hoursDimensions (W/H/D)178.5 x 247.6 x 6.1mm214.9 x 280.6 x 6.1mmWeight460g (Wi-Fi); 460g (cellular)616g (Wi-Fi); 617g (cellular)Table: Thomas Armbrüster

In my experience, most UX teams find themselves primarily implementing other people’s ideas rather than leading the conversation about user experience. This happens because stakeholders and decision-makers often lack a deep understanding of UX’s capabilities and potential. Without a clear UX strategy framework, professionals get relegated to a purely tactical role — wireframing and testing solutions conceived by others. A well-crafted UX strategy framework changes this dynamic. It helps UX teams take control of their role and demonstrate real leadership in improving the user experience. Rather than just responding to requests, you can proactively identify opportunities that deliver genuine business value. A strategic approach also helps educate stakeholders about UX’s full potential while building credibility through measurable results. Strategy And The Fat Smoker When I guide teams on creating a UX strategy, I like to keep things simple. I borrow an approach from the book Strategy and the Fat Smoker and break strategy into three clear parts: First, we diagnose where we are today. Then, we set guiding policies to steer us. Finally, we outline actions to get us where we want to go. Let me walk you through each part so you can shape a UX strategy that feels both practical and powerful. Diagnosis: Know Your Starting Point Before we outline any plan, we need to assess our current situation. A clear diagnosis shows where you can make the biggest impact. It also highlights the gaps you must fill. Identify Status Quo Failures Start by naming what isn’t working. You might find that your organization lacks a UX team. Or the team has a budget that is too small. Sometimes you uncover that user satisfaction scores are slipping. Frame these challenges in business terms. For example, a slow sign‑up flow may be costing you 20 percent of new registrations each month. That ties UX to revenue and grabs attention. Once you have a list of failures, ask yourself: What outcome does each failure hurt? A slow checkout might reduce e‑commerce sales. Complicated navigation may dent customer retention. Linking UX issues to business metrics makes the case for change. Map The Aspirational Experience Next, visualize what an improved journey would look like. A quick way is to create two simple journey maps. One shows the current experience. The other shows an ideal path. Highlight key steps like discovery, sign‑up, onboarding, and support. Then ask: How will this new journey help meet our business goals? Maybe faster onboarding can cut support costs. Or a streamlined checkout can boost average order value. Let me share a real-world example. When working with the Samaritans, a UK mental health charity, we first mapped their current support process. While their telephone support was excellent, they struggled with email and text support, and had no presence on social media platforms. This was largely because volunteers found it difficult to manage multiple communication systems. We then created an aspirational journey map showing a unified system where volunteers could manage all communication channels through a single interface. This clear vision gave the organization a concrete goal that would improve the experience for both users seeking help and the volunteers providing support. This vision gives everyone something to rally around. It also guides your later actions by showing the target state. Audit Resources And Influence Next, turn your attention to what you have to work with. List your UX team members and their skills. Note any budget set aside for research tools or software licenses. Then identify where you have influence across the organization. Which teams already seek your advice? Who trusts your guidance? That might be the product group or marketing. You’ll lean on these allies to spread UX best practices. Finally, consider who else matters. Are there policy owners, process leads, or executives you need on board? Jot down names and roles so you can loop them in later. Spot Your Constraints Every strategy must live within real‑world limits. Maybe there’s a headcount freeze. Or IT systems won’t support a major overhaul. List any technical, budget, or policy limits you face. Then accept them. You’ll design your strategy to deliver value without asking for impossible changes. Working within constraints boosts your credibility. It also forces creativity. With the diagnosis complete, we know where we stand. Next, let’s look at how to steer our efforts. Guiding Policies: Set the North Star Guiding policies give you guardrails. They help you decide which opportunities to chase and which to skip. These policies reflect your priorities and the best path forward. Choose A Tactical Or Strategic Approach Early on, you must pick how your UX team will operate. You have two broad options: TacticalYou embed UX people on specific projects. They run tests and design interfaces hands‑on. This needs a bigger team. I like a ratio of one UX pro for every two developers. StrategicYou act as a center of excellence. You advise other teams. You build guidelines, run workshops, and offer tools. This needs fewer hands but a broader influence. Weigh your resources against your goals. If you need to move fast on many projects, go tactical. If you want to shift mindsets, work strategically. Choose the approach with the best chance of success. Define A Prioritization Method You’ll face many requests for UX work. A clear way to sort them saves headaches. Over the years, I’ve used a simple digital triage. You score each request based on impact, effort, and risk. Then, you work on the highest‑scoring items first. You can adapt this model however you like. The point is to have a repeatable, fair way to say yes or no. Create A Playbook Of Principles A playbook holds your core design principles, standard operating procedures, and templates. It might include: A design system for UI patterns; Standards around accessibility or user research; Guides for key tasks such as writing for the web; Templates for common activities like user interviews. This playbook becomes your team’s shared reference. It helps others repeat your process. It also captures the know‑how you need as your team grows. Plan Your Communication Strategy fails when people don’t know about it. You need a plan to engage stakeholders. I find it helpful to use a RACI chart — who is Responsible, Accountable, Consulted, and Informed. Then decide: How often will you send updates? Which channels should you use (email, Slack, weekly demos)? Who leads each conversation? Clear, regular communication keeps everyone looped in. It also surfaces concerns early so you can address them. With guiding policies in place, you have a clear way to decide what to work on. Now, let’s turn to making things happen. Action Plan: Bring Strategy To Life Actions are the concrete steps you take to deliver on your guiding policies. They cover the projects you run, the support you give, and the risks you manage. Outline Key Projects And Services Start by listing the projects you’ll tackle. These might be: Running a discovery phase for a new product. Building a design system for your marketing team. Conducting user tests on your main flow. For each project, note what you will deliver and when. You can use your digital triage scores to pick the highest priorities. Keep each project scope small enough to finish in a few sprints. That way, you prove value quickly. Offer Training And Tools If you choose a strategic approach, you need to empower others. Plan workshops on core UX topics. Record short videos on testing best practices. Build quick reference guides. Curate a list of tools: Prototyping apps, Research platforms, Analytics dashboards. Make these resources easy to find in your playbook. Assign Stakeholder Roles Your strategy needs executive backing. Identify a senior sponsor who can break through roadblocks. Outline what you need them to do. Maybe it’s championing a new budget line or approving key hires. Also, pin down other collaborators. Who on the product side will help you scope new features? Who on the IT team will support user research tooling? Getting clear roles avoids confusion. Manage Risks and Barriers No plan goes off without a hitch. List your biggest risks, such as: A hiring freeze delays tactical hires; Key stakeholders lose interest; Technical debt slows down new releases. For each risk, jot down how you’ll handle it. Maybe you should shift to a fully strategic approach if hiring stalls. Or you can send a weekly one‑page update to reengage sponsors. Having a fallback keeps you calm when things go sideways. Before we wrap up, let’s talk about making strategy stick. Embedding UX Into The Culture A strategy shines only if you deeply embed it into your organization’s culture. Here’s how to make that happen: Build awareness and enthusiasm Run regular “lunch and learn” sessions to showcase UX wins. Host an annual UX day or mini-conference to boost visibility. Create a monthly UX salon where teams share challenges and victories. Make UX visible and tangible Display personas and journey maps in office spaces. Add design principles to everyday items like mousepads and mugs. Share success metrics and improvements in company communications. Embed UX into processes Establish clear UX policies and best practices. Review and update procedures that might hinder a good user experience. Create a healthy competition between teams through UX metrics. These tactics transform your strategy from a document into an organizational movement. They foster a culture where everyone thinks about user experience, not just the UX team. Remember, cultural change takes time — but consistent, visible efforts will gradually shift mindsets across the organization. Implementing Your UX Strategy: From Plan To Practice We started by diagnosing your current state. Then we set policies to guide your efforts. Finally, we laid out an action plan to deliver results. This three-part framework keeps your UX work tied to real business needs. It also gives you clarity, focus, and credibility. However, creating a strategy is the easy part — implementing it is where the real challenge lies. This is precisely why the book Strategy and the Fat Smoker carries its distinctive title. Just as someone who is overweight or smokes knows exactly what they need to do, we often know what our UX strategy should be. The difficult part is following through and making it a reality. Success requires consistent engagement and persistence in the face of setbacks. As Winston Churchill wisely noted, “Success is going from failure to failure with no loss of enthusiasm.” This perfectly captures the mindset needed to implement a successful UX strategy — staying committed to your vision even when faced with obstacles and setbacks.

European Union App Store users are being shown warnings in the App Store about apps that let users make payments directly to developers — despite this being a clear violation of the Digital Markets Act.Apple has already been fined $570 million for anticompetitive practices under the EU's Digital Markets Act (DMA), although it is appealing. Now it at least appears to be asking for a further fine, because it has added warnings expressly designed to dissuade users from buying apps with alternative payment processes.It's actually in the recent US ruling that Apple was explicitly forbidden to use what are called "scare screens". The EU has not as yet ruled against that specific technique, although in March 2024 it was alerted to Apple's plans by rivals including Epic Games.Nonetheless, the EU has focused on Apple's alleged anticompetitive practices. In its announcement of its ruling in April 2025, the EU said that it "ordered Apple to remove the technical and commercial restrictions on steering and to refrain from perpetuating the non-compliant conduct in the future, which includes adopting conduct with an equivalent object or effect." Continue Reading on AppleInsider | Discuss on our Forums

Hideo Kojima, known for his unconventional approach to game design, does not seem to believe in crowd-pleasers. Despite that, Death Stranding 2: On the Beach might become a mainstream hit. VGC reported that in a recent interview with Edge Magazine Kojima said, “I do wish I was a bit more controversial,” reflecting on the positive reaction Death Stranding 2 received from internal playtesters. He contrasted the internal reception for Death Stranding 2 to the original Death Stranding: “In fact, I did a lot of monitoring tests for the original Death Stranding. I’d find that four out of ten people would love the game, but six would say it was a terrible game. For me, I think that’s a good balance. But with the playtests for Death Stranding 2: On the Beach, everyone seems to be positive.” Edge 411 is out today, featuring Death Stranding 2: On The Beach. For our 16-page cover story, Simon Parkin visits Kojima Productions to play the studio’s most ambitious game to date and discuss its creation with Hideo Kojima, who explains why it is more Aliens than Alien. pic.twitter.com/AUyceJyBOA— Edge (@edgeonline) May 15, 2025 That type of pre-release praise is normally music to a game studio founder’s ears, but Kojima is different. “I’m not interested in making something that appeals to everyone,” he said in the Edge interview. While the original Death Stranding — with its bold package delivery premise — was not a hit right out of the gate, more critics came around to it with its Director’s Cut edition. Amid a warming reception to the original Death Stranding, it makes sense that play testers for Death Stranding 2 are more ready to buy in. Well, there’s that improving climate, and there’s also the fact that Death Stranding 2 looks like it’s building on the first game’s already ambitious base. Natural disasters like earthquakes and sandstorms create new challenges for traversing the world. On top of highways, players can also now set up efficient, but expensive, trade routes through monorails. Combat is becoming more core to the experience, with a new arsenal of sniper rifles, assault rifles, grenade launchers, and more. Even Sam’s backpack can be put down, allowing for more nimble combat and infiltration sequences. All together, it looks like Death Stranding 2 promises more freedom in how to overcome obstacles. With all that packed inside an already unforgettable world, it is no surprise that play testers are responding positively. Balancing the desire for originality with more widespread success is a key consideration for Kojima, he said. “Of course, I can’t afford to make a failure, either. I must take care of my studio, my staff, and our production budget. But in essence, I don’t want to simply recreate something that’s already in the world,” Kojima said. “I want to make something new.” Death Stranding 2: On the Beach is set to be released June 26, 2025 on PlayStation 5.

Text-to-audio generation has emerged as a transformative approach for synthesizing sound directly from textual prompts, offering practical use in music production, gaming, and virtual experiences. Under the hood, these models typically employ Gaussian flow-based techniques such as diffusion or rectified flows. These methods model the incremental steps that transition from random noise to structured audio. While highly effective in producing high-quality soundscapes, the slow inference speeds have posed a barrier to real-time interactivity. It is particularly limiting when creative users expect an instrument-like responsiveness from these tools. Latency is the primary issue with these systems. Current text-to-audio models can take several seconds or even minutes to generate a few seconds of audio. The core bottleneck lies in their step-based inference architecture, requiring between 50 and 100 iterations per output. Previous acceleration strategies focus on distillation methods where smaller models are trained under the supervision of larger teacher models to replicate multi-step inference in fewer steps. However, these distillation methods are computationally expensive. They demand large-scale storage for intermediate training outputs or require simultaneous operation of several models in memory, which hinders their adoption, especially on mobile or edge devices. Also, such methods often sacrifice output diversity and introduce over-saturation artifacts. While a few adversarial post-training methods have been attempted to bypass the cost of distillation, their success has been limited. Most existing implementations rely on partial distillation for initialization or do not scale well to complex audio synthesis. Also, audio applications have seen fewer fully adversarial solutions. Tools like Presto integrate adversarial objectives but still depend on teacher models and CFG-based training for prompt adherence, which restricts their generative diversity. Researchers from UC San Diego, Stability AI, and Arm introduced Adversarial Relativistic-Contrastive (ARC) post-training. This approach sidesteps the need for teacher models, distillation, or classifier-free guidance. Instead, ARC enhances an existing pre-trained rectified flow generator by integrating two novel training objectives: a relativistic adversarial loss and a contrastive discriminator loss. These help the generator produce high-fidelity audio in fewer steps while maintaining strong alignment with text prompts. When paired with the Stable Audio Open (SAO) framework, the result was a system capable of generating 12 seconds of 44.1 kHz stereo audio in only 75 milliseconds on an H100 GPU and around 7 seconds on mobile devices. With ARC methodology, they introducedStable Audio Open Small, a compact and efficient version of SAO tailored for resource-constrained environments. This model contains 497 million parameters and uses an architecture built on a latent diffusion transformer. It consists of three main components: a waveform-compressing autoencoder, a T5-based text embedding system for semantic conditioning, and a DiT (Diffusion Transformer) that operates within the latent space of the autoencoder. Stable Audio Open Small can generate stereo audio up to 11 seconds long at 44.1 kHz. It is designed to be deployed using the ‘stable-audio-tools’ library and supports ping-pong sampling, enabling efficient few-step generation. The model demonstrated exceptional inference efficiency, achieving generation speeds of under 7 seconds on a Vivo X200 Pro phone after applying dynamic Int8 quantization, which also cut RAM usage from 6.5GB to 3.6 GB. This makes it especially viable for on-device creative applications like mobile audio tools and embedded systems. The ARC training approach involves replacing the traditional L2 loss with an adversarial formulation where generated and real samples, paired with identical prompts, are evaluated by a discriminator trained to distinguish between them. A contrastive objective teaches the discriminator to rank accurate audio-text pairs higher than mismatched ones to improve prompt relevance. These paired objectives eliminate the need for CFG while achieving better prompt adherence. Also, ARC adopts ping-pong sampling to refine the audio output through alternating denoising and re-noising cycles, reducing inference steps without compromising quality. ARC’s performance was evaluated extensively. In objective tests, it achieved an FDopenl3 score of 84.43, a KLpasst score of 2.24, and a CLAP score of 0.27, indicating balanced quality and semantic precision. Diversity was notably strong, with a CLAP Conditional Diversity Score (CCDS) of 0.41. Real-Time Factor reached 156.42, reflecting outstanding generation speed, while GPU memory usage remained at a practical 4.06 GB. Subjectively, ARC scored 4.4 for diversity, 4.2 for quality, and 4.2 for prompt adherence in human evaluations involving 14 participants. Unlike distillation-based models like Presto, which scored higher on quality but dropped to 2.7 on diversity, ARC presented a more balanced and practical solution. Several Key Takeaways from the Research by Stability AI on Adversarial Relativistic-Contrastive (ARC) post-training and  Stable Audio Open Small include:  ARC post-training avoids distillation and CFG, relying on adversarial and contrastive losses. ARC generates 12s of 44.1 kHz stereo audio in 75ms on H100 and 7s on mobile CPUs. It achieves 0.41 CLAP Conditional Diversity Score, the highest among tested models. Subjective scores: 4.4 (diversity), 4.2 (quality), and 4.2 (prompt adherence). Ping-pong sampling enables few-step inference while refining output quality. Stable Audio Open Small offers 497M parameters, supports 8-step generation, and is compatible with mobile deployments. On Vivo X200 Pro, inference latency dropped from 15.3s to 6.6s with half the memory. ARC and SAO Small provide real-time solutions for music, games, and creative tools. In conclusion, the combination of ARC post-training and Stable Audio Open Small eliminates the reliance on resource-intensive distillation and classifier-free guidance, enabling researchers to deliver a streamlined adversarial framework that accelerates inference without compromising output quality or prompt adherence. ARC enables fast, diverse, and semantically rich audio synthesis in high-performance and mobile environments. With Stable Audio Open Small optimized for lightweight deployment, this research lays the groundwork for integrating responsive, generative audio tools into everyday creative workflows, from professional sound design to real-time applications on edge devices. Check out the Paper, GitHub Page and Model on Hugging Face. 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 90k+ ML SubReddit. Mohammad AsjadAsjad is an intern consultant at Marktechpost. He is persuing B.Tech in mechanical engineering at the Indian Institute of Technology, Kharagpur. Asjad is a Machine learning and deep learning enthusiast who is always researching the applications of machine learning in healthcare.Mohammad Asjadhttps://www.marktechpost.com/author/mohammad_asjad/Meta AI Introduces CATransformers: A Carbon-Aware Machine Learning Framework to Co-Optimize AI Models and Hardware for Sustainable Edge DeploymentMohammad Asjadhttps://www.marktechpost.com/author/mohammad_asjad/Enterprise AI Without GPU Burn: Salesforce’s xGen-small Optimizes for Context, Cost, and PrivacyMohammad Asjadhttps://www.marktechpost.com/author/mohammad_asjad/ServiceNow AI Released Apriel-Nemotron-15b-Thinker: A Compact Yet Powerful Reasoning Model Optimized for Enterprise-Scale Deployment and EfficiencyMohammad Asjadhttps://www.marktechpost.com/author/mohammad_asjad/Researchers from Fudan University Introduce Lorsa: A Sparse Attention Mechanism That Recovers Atomic Attention Units Hidden in Transformer Superposition