XAI LAUNCHES AN API FOR GENERATING IMAGES
xAI's grok-2-image-1212 can generate up to 10 images per request (limited to five requests per second) in JPG format for $0.07 per image.
Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
xAI's grok-2-image-1212 can generate up to 10 images per request (limited to five requests per second) in JPG format for $0.07 per image.
The future of no-code isn't better visual programming tools, it's a world where writing code is as easy as having a conversation.
A list of things you can do to live a secure digital life.
A list of coding blindspots in large language models (focused on the Sonnet family).
FutureHouse is a nonprofit determined to automate scientific discovery using AI agents.
Sponsored
Manual testing on personal devices is TOO SLOW AND LIMITED. Teams CUT RELEASES A WEEK EARLY just to test before submitting them to app stores— and without broad device coverage, issues slip through.
Nvidia has announced new chips for building and deploying artificial intelligence models. Blackwell Ultra will allow cloud providers to make as much as 50 times the revenue as chips from the Hopper generation - it will ship in the second half of this year. Vera Rubin, which is expected to ship in the second half of 2026, is a system with two main components: a CPU called Vera and a new GPU design called Rubin. The custom Vera design will be twice as fast as the CPU used in last year's Grace Blackwell chips - it will enable Rubin to manage 50 petaflops while doing inference, more than double the capacity of the company's current Blackwell chips.
Google is acquiring Wiz for $32 billion. Wiz will continue to be positioned as a multicloud offering. The acquisition brings a massive customer list to Google, so the companies need to ensure existing customers don't start shopping around for another security provider. Many customers chose Wiz due to its ability to support multiple cloud platforms - cutting off that ability risks alienating those users. Google Cloud trails significantly behind its competitors with a 12% share of the global cloud market.
Nvidia and Google DeepMind's Newton is a physics engine that can simulate robotic movements in real-world settings. It will be used by Disney to power its next-generation entertainment robots - a video of one of these robots is available in the article. Newton helps robots be more expressive and learn how to handle complex tasks with precision. It is highly customizable - developers can use it to program robotic interactions with a range of deformable objects. Newton will be compatible with Google DeepMind's ecosystem of robotic development tools.
Firefly Aerospace's Blue Ghost lander, part of NASA's Commercial Lunar Payload Services (CLPS) program, has accomplished a lot on the Moon in the last two weeks. It has drilled into the Moon's surface, tested an extraterrestrial vacuum cleaner, demonstrated that GPS can be used to navigate on the lunar surface, and sent daily imagery back to Earth. All of the mission objectives for the Blue Ghost mission have been met. This article details how Firefly managed to secure the first fully successful commercial lunar landing, the challenges Firefly's engineers have had to overcome, some of the discoveries that Blue Ghost has made so far, and the progress of other missions in the CLPS program.
On-call, in its current state in Big Tech, is flawed. AI is already changing the scope of on-call and the responsibility of the engineer in the loop. Some on-call will always be unavoidable, but it should be the exception rather than the norm. This will give engineers time to concentrate on meaningful tasks, which will make customers happier.
Services should minimize the number of synchronous requests they make to other services as much as possible. Synchronous calls are costly and impact the availability of services. The more services that an app relies on synchronously, the lower the availability of that app will be. Making a synchrony budget is about being mindful of how communication flows are implemented between services.
Keith O'Brien, a global payroll-compliance manager at Rippling, a business software startup, has been accused of stealing trade secrets for a rival, Deel, as part of a spying operation. O'Brien allegedly searched for 'deel' an average of 23 times a day across Rippling's internal systems, accessing message boards that had nothing to do with his job in a mode that hid his presence from others. Rippling claims that it set a honeypot trap that proved O'Brien was being directed by Deel's top executives. O'Brien fleed Rippling's offices with his smartphone after being accused of being a corporate spy despite being told that failing to turn over the device violated a court order and could result in imprisonment.
Core Devices, a company by the original creator of the Pebble smartwatch, is getting ready to ship two new watches. They will have all of the features of the old Pebble devices and run the newly open-sourced Pebble operating system. The Core 2 Duo will ship in July for $149. It has the exact same black-and-white e-paper display as the old Pebble 2 and even comes in the exact same frame. It has a battery that lasts more than 30 days and an in-device speaker for chatting with AI assistants. The Core Time 2 will ship in December and cost $225 - it will have a touchscreen.
NVIDIA Dynamo is an inference engine agnostic, high-throughput, low-latency inference framework designed for serving generative AI and reasoning models in multi-node distributed environments.
The construction for the site, which will have a capacity of 1.2 gigawatts of power, will be completed by mid-2026.
EBS failure is constant, variable, and all by design.
CoreWeave has filed paperwork for a public listing - if the initial public offering goes ahead, it will be the first prominent AI startup to reach the stock market.
This post shows how to increase jq's performance by 90% by just compiling the package again using the same source package that Ubuntu uses.
Llama has achieved widespread success since launching in 2023, with companies including Spotify, AT&T, and DoorDash using Llama models in production today.
Sponsored
Modern bots are smarter than ever [5]—executing JavaScript, storing cookies, rotating IPs, and even cracking CAPTCHAs with AI. As attacks grow more sophisticated, traditional detection just isn't enough.
OpenAI plans to start testing ways for business customers to connect apps to ChatGPT. It will soon begin beta testing a new feature called ChatGPT Connectors that will allow ChatGPT Team subscribers to link workspace Google Drive and Slack accounts to ChatGPT so the chatbot can answer questions informed by content across those accounts. None of the data synced from the apps will be used for training. The company plans to expand the feature to other platforms like Microsoft SharePoint and Box in the future.
Apple plans to launch its first foldable in 2026. Market expectations for iPhone sales are lower due to the delay of the new Siri experience, but the launch of a foldable may help the company recover from weak demand for iPhones. At least one new foldable is targeted to enter mass production in the second half of 2026. The first foldable iPhone may cost more than $2,000 and look like an iPad mini when unfolded. Apple is also working on a foldable device with an 18-inch screen that could end up becoming a new Mac or iPad.
BYD has a new EV architecture that will enable charging parity with gas station visits. The Super E-Platform promises 1,000 kW charge speeds - double the current industry leaders. Most fast chargers on the market can only reach about 350 kW and many of them are usually more in the 200s. BYD has launched new Han L and Tang L models that use the new fast-charging platform. The company plans to implement over 4,000 ultra-fast charging stations around China.
A team of Korean researchers has engineered a bacterial strain that can make useful polymers starting with nothing but glucose for fuel. The system can be tweaked to incorporate a huge range of chemicals into a variety of polymers. The resulting plastic can be tuned across many properties. There are still some issues, as the process doesn't allow complete control over what gets added into the polymer - the scientists couldn't stop enzymes from mixing random chemicals from the cells' metabolism into the polymers - and the polymers still need to be purified from the rest of the cell components before being used in manufacturing.
The new Discord Social SDK is a free toolkit that allows developers to enhance their games' social and multiplayer experiences by leveraging Discord's social infrastructure. Developers will be able to improve their in-game experiences for players regardless of whether they have a Discord account. The social integrations include a unified friends list and deep-linked game invites that enable players to invite their friends to directly join their party or lobby. The SDK is compatible with C++, Unreal Engine, and Unity and supports Windows 11+ and macOS - support for console and mobile is coming.
There are many beliefs about the limits, usefulness, and potential of large language models. Every piece of software needs to be 'correct', which means that the software conforms to the intent of the creator. The creator has to verify such conformance - it cannot be offloaded. Verifiability is the limit that tells us what we can't implement via large language models, and it cannot be solved with agentic approaches.
HR company Rippling has launched a lawsuit against Deel, another big player in the industry, alleging racketeering, misappropriation of trade secrets, tortious interference, unfair competition, and aiding and abetting a breach of fiduciary duty. Deel has denied the allegations, saying the lawsuit was an attempt to shift attention away from Rippling after it was accused of violating sanctions law in Russia. A lot of the evidence for Rippling's claims is based on Slack activity.
Generative AI chat interfaces are changing how consumers act online. Adobe has issued a report that shows how generative AI sources are directing a significant surge of traffic to US websites. Generative AI has been embraced by users as a shopping assistant. Shoppers say AI-powered chat interfaces help shorten the time required for them to receive information that is personalized to their needs. The technology is causing businesses to rethink customer engagement, especially as future AI agents will be able to handle more complex tasks and make highly tailored recommendations.
The model can also reportedly generate celebrity images.
Ejectable apps allow users to 'eject' at any time and switch to self-hosting (or vice versa).
xAI has acquired Hotshot, a startup working on AI-powered video generation tools similar to OpenAI's Sora.
Amazon's devices chief Panos Panay is adding a premium tier to the next generation of artificially intelligent Alexa gadgets.
CloudKitchens appears to be aiming to eventually deliver AI-perfected meals directly to customers.
Alphabet is in advanced talks to acquire Wiz at a price of around $30 billion.
Sponsored
EmpowerHR North America is just THREE DAYS AWAY—but there's still time to grab your free spot [4]. Here's why you should:
Robby Walker, the executive overseeing Apple's Siri revamp, said that the delays to key features have been ugly and embarrassing and that the decision to publicly promote the technology before it was ready made matters worse. It is unknown when the enhancements will actually launch, but they aren't expected until next year at the earliest. These features were a key selling point of the iPhone 16 line, which otherwise lacked major changes. Apple reportedly doesn't plan to immediately fire any top executives over the AI crisis, but the company is poised to make management adjustments.
Astronauts Butch Wilmore and Suni Williams, who were only supposed to be on the International Space Station for eight days, will begin their journey back to Earth later this week. A SpaceX capsule carrying a new crew has docked at the ISS. There will be a two-day handover period after which the old crew will begin their journey - there could be a small further delay due to weather conditions. The astronauts have consistently said that they have been happy to be on board the space station.
Intuitive Machines' Athena spacecraft's altimeter failed during landing last week, causing it to misjudge its altitude and land in a small shadowed crater after toppling over and rotating once or twice. This caused regolith to be blown up into the solar panels, which meant the panels weren't able to provide the lander sufficient power to survive. The lander was able to accomplish some of its objectives as mission operators decided to operate as robustly as possible for 13 hours to get as much data as they could. Despite the bad landing, the mission was largely a success - Athena flew smoothly, the company validated the spacecraft's methane-based propulsion system, and it also proved that its communications technology can be used as part of the lunar data relay network that NASA has contracted with Intuitive Machines to develop.
The past year has seen some really interesting feature launches for S3 that remove limits and make S3 simpler. A lot of work has gone into the properties that make S3 simple - the S3 team considers themselves successful only when these things can be taken for granted as that means they aren't distractions. All of the most popular features on S3 have been in direct response to requests from S3 customers - this article looks at how the S3 team has evolved its product in response to the huge range of things that builders do today while keeping it simple so that developers can work with data and not have to think about anything else.
Excellent project management is an extremely high-leverage skill - organizing a ton of moving parts with complex, non-obvious interdependencies and hard schedule constraints is a huge job and can save weeks of delays if done right. Project management requires focused time for organization. A detailed plan for victory is important because whether or not the plan is being achieved is the best way to figure out how well or badly things are going. Getting complete information can often be the hardest part of the project. It's important to overcommunicate so everyone knows what else is going on to make correct decisions. Delegate when the project gets too big - sometimes it's impossible to track every detail by yourself.
The latest cohort of Y Combinator startups is growing significantly faster than past cohorts - and with actual revenue. The entire batch in aggregate grew 10% per week over the last nine months. Y Combinator CEO Garry Tan attributes the growth spurt to leaps in artificial intelligence. AI's ability to subsidize an otherwise heavy workload has allowed these companies to build with fewer people. 95% of the code was written by AI for about a quarter of the current Y Combinator startups. While there's a ton of hype, the startups are actually getting commercial validation.
Being a software engineer has been a lot of fun for most of the last decade, with good job opportunities and plenty of perks for working in tech. That has changed in the last two years. A lot of software engineers are refusing to change - there is going to be a personal cost to anyone not going along with the vibe shift, particularly for more junior or more vulnerable engineers. Companies are now trying to focus, so funding and resources are more focused on a handful of bets that company executives actually care about. The silver lining is that the industry is now more realistic and value is easier to understand.
Developers can now experiment and even distribute APKs using the Meta Quest's new Passthrough Camera API, but they can't use it yet in Meta Horizon Store apps.
If you are finding the job market difficult at the moment, it's not personal - those who are in frustrating roles should try harder than usual to find a way to make it a rewarding experience.
Tesla plans to use an internal fleet backed by teleoperation support in a geo-fenced and mapped area when it launches its robotaxi service in Austin, Texas, in June.
Large language models' ability to compress information, even if imprecise, can help preserve pieces of the Internet that would otherwise be lost to time.