Meta released two new Llama 4 models over the weekend -- Scout and Maverick -- with claims that Maverick outperforms GPT-4o and Gemini 2.0 Flash on benchmarks. Maverick quickly secured the number-two spot on LMArena, behind only Gemini 2.5 Pro.

Researchers have since discovered that Meta used an "experimental chat version" of Maverick for LMArena testing that was "optimized for conversationality" rather than the publicly available version.

In response, LMArena said "Meta's interpretation of our policy did not match what we expect from model providers" and announced policy updates to prevent similar issues.

A court has blocked a British government attempt to keep secret a legal case over its demand to access Apple user data. From a report: The UK Investigatory Powers Tribunal, a special court that handles cases related to government surveillance, said the authorities' efforts were a "fundamental interference with the principle of open justice" in a ruling issued on Monday. The development comes after it emerged in January that the British government had served Apple with a demand to circumvent encryption that the company uses to secure user data stored in its cloud services.

Apple challenged the request, while taking the unprecedented step of removing its advanced data protection feature for its British users. The government had sought to keep details about the demand -- and Apple's challenge of it -- from being publicly disclosed. Apple has regularly clashed with governments over encryption features that can make it difficult for law enforcement to access devices produced by the company. The world's most valuable company last year criticized UK surveillance powers as "unprecedented overreach" by the government.

It's "a milestone for Meta AI and for open source," Mark Zuckerberg said this weekend. "For the first time, the best small, mid-size, and potentially soon frontier [large-language] models will be open source."

Zuckerberg anounced four new Llama LLMs in a video posted on Instagram and Facebook — two dropping this weekend, with another two on the way. "Our goal is to build the world's leading AI, open source it, and make it universally accessible so that everyone in the world benefits."

Zuckerberg's announcement: I've said for a while that I think open source AI is going to become the leading models. And with Llama 4 this is starting to happen.

- The first model is Llama 4 Scout. It is extremely fast, natively multi-modal. It has an industry-leading "nearly infinite" 10M-token context length, and is designed to run on a single GPU. [Meta's blog post says it fits on an NVIDIA H100]. It is 17 billion parameters by 16 experts, and it is by far the highest performing small model in its class.

- The second model is Llama 4 Maverick — the workhorse. It beats GPT-4o and Gemini Flash 2 on all benchmarks. It is smaller and more efficient than DeepSeek v3, but it is still comparable on text, plus it is natively multi-modal. This one is 17B parameters x 128 experts, and it is designed to run on a single host for easy inference.

This thing is a beast.
Zuck promised more news next month on "Llama 4 Reasoning" — but the fourth model will be called Llama 4 Behemoth. "This thing is massive. More than 2 trillion parameters." (A blog post from Meta AI says it also has a 288 billion active parameter model, outperforms GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on STEM benchmarks, and will "serve as a teacher for our new models.")

"I'm not aware of anyone training a larger model out there," Zuckberg says in his video, calling Behemoth "already the highest performing base model in the world, and it is not even done training yet."

"If you want to try Llama 4, you can use Meta AI in WhatsApp, Messenger, or Instagram Direct," Zuckberg said in his video, "or you can go to our web site at meta.ai." The Scout and Maverick models can be downloaded from llama.com and Hugging Face.

"We continue to believe that openness drives innovation," Meta AI says in their blog post, "and is good for developers, good for Meta, and good for the world." Their blog post declares it's "The beginning of a new era of natively multimodal AI innovation," calling Scout and Maverick "the best choices for adding next-generation intelligence." This is just the beginning for the Llama 4 collection. We believe that the most intelligent systems need to be capable of taking generalized actions, conversing naturally with humans, and working through challenging problems they haven't seen before. Giving Llama superpowers in these areas will lead to better products for people on our platforms and more opportunities for developers to innovate on the next big consumer and business use cases. We're continuing to research and prototype both models and products, and we'll share more about our vision at LlamaCon on April 29...

We also can't wait to see the incredible new experiences the community builds with our new Llama 4 models.
"The impressive part about Llama 4 Maverick is that with just 17B active parameters, it has scored an ELO score of 1,417 on the LMArena leaderboard," notes the tech news site Beebom. "This puts the Maverick model in the second spot, just below Gemini 2.5 Pro, and above Grok 3, GPT-4o, GPT-4.5, and more.

"It also achieves comparable results when compared to the latest DeepSeek V3 model on reasoning and coding tasks, and surprisingly, with just half the active parameters."

Slashdot reader zlives shared this report from BleepingComputer: Microsoft used its AI-powered Security Copilot to discover 20 previously unknown vulnerabilities in the GRUB2, U-Boot, and Barebox open-source bootloaders.

GRUB2 (GRand Unified Bootloader) is the default boot loader for most Linux distributions, including Ubuntu, while U-Boot and Barebox are commonly used in embedded and IoT devices. Microsoft discovered eleven vulnerabilities in GRUB2, including integer and buffer overflows in filesystem parsers, command flaws, and a side-channel in cryptographic comparison. Additionally, 9 buffer overflows in parsing SquashFS, EXT4, CramFS, JFFS2, and symlinks were discovered in U-Boot and Barebox, which require physical access to exploit.

The newly discovered flaws impact devices relying on UEFI Secure Boot, and if the right conditions are met, attackers can bypass security protections to execute arbitrary code on the device. While exploiting these flaws would likely need local access to devices, previous bootkit attacks like BlackLotus achieved this through malware infections.
Miccrosoft titled its blog post "Analyzing open-source bootloaders: Finding vulnerabilities faster with AI." (And they do note that Micxrosoft disclosed the discovered vulnerabilities to the GRUB2, U-boot, and Barebox maintainers and "worked with the GRUB2 maintainers to contribute fixes... GRUB2 maintainers released security updates on February 18, 2025, and both the U-boot and Barebox maintainers released updates on February 19, 2025.")

They add that performing their initial research, using Security Copilot "saved our team approximately a week's worth of time," Microsoft writes, "that would have otherwise been spent manually reviewing the content." Through a series of prompts, we identified and refined security issues, ultimately uncovering an exploitable integer overflow vulnerability. Copilot also assisted in finding similar patterns in other files, ensuring comprehensive coverage and validation of our findings...

As AI continues to emerge as a key tool in the cybersecurity community, Microsoft emphasizes the importance of vendors and researchers maintaining their focus on information sharing. This approach ensures that AI's advantages in rapid vulnerability discovery, remediation, and accelerated security operations can effectively counter malicious actors' attempts to use AI to scale common attack tactics, techniques, and procedures (TTPs).
This week Google also announced Sec-Gemini v1, "a new experimental AI model focused on advancing cybersecurity AI frontiers."

The advent of LLMs and machine learning-based applications "opened the door to a new wave of security threats," argues Google's security blog. (Including model and data poisoning, prompt injection, prompt leaking and prompt evasion.)

So as part of the Linux Foundation's nonprofit Open Source Security Foundation, and in partnership with NVIDIA and HiddenLayer, Google's Open Source Security Team on Friday announced the first stable model-signing library (hosted at PyPI.org), with digital signatures letting users verify that the model used by their application "is exactly the model that was created by the developers," according to a post on Google's security blog. [S]ince models are an uninspectable collection of weights (sometimes also with arbitrary code), an attacker can tamper with them and achieve significant impact to those using the models. Users, developers, and practitioners need to examine an important question during their risk assessment process: "can I trust this model?"

Since its launch, Google's Secure AI Framework (SAIF) has created guidance and technical solutions for creating AI applications that users can trust. A first step in achieving trust in the model is to permit users to verify its integrity and provenance, to prevent tampering across all processes from training to usage, via cryptographic signing... [T]he signature would have to be verified when the model gets uploaded to a model hub, when the model gets selected to be deployed into an application (embedded or via remote APIs) and when the model is used as an intermediary during another training run. Assuming the training infrastructure is trustworthy and not compromised, this approach guarantees that each model user can trust the model...

The average developer, however, would not want to manage keys and rotate them on compromise. These challenges are addressed by using Sigstore, a collection of tools and services that make code signing secure and easy. By binding an OpenID Connect token to a workload or developer identity, Sigstore alleviates the need to manage or rotate long-lived secrets. Furthermore, signing is made transparent so signatures over malicious artifacts could be audited in a public transparency log, by anyone. This ensures that split-view attacks are not possible, so any user would get the exact same model. These features are why we recommend Sigstore's signing mechanism as the default approach for signing ML models.

Today the OSS community is releasing the v1.0 stable version of our model signing library as a Python package supporting Sigstore and traditional signing methods. This model signing library is specialized to handle the sheer scale of ML models (which are usually much larger than traditional software components), and handles signing models represented as a directory tree. The package provides CLI utilities so that users can sign and verify model signatures for individual models. The package can also be used as a library which we plan to incorporate directly into model hub upload flows as well as into ML frameworks.
"We can view model signing as establishing the foundation of trust in the ML ecosystem..." the post concludes (adding "We envision extending this approach to also include datasets and other ML-related artifacts.") Then, we plan to build on top of signatures, towards fully tamper-proof metadata records, that can be read by both humans and machines. This has the potential to automate a significant fraction of the work needed to perform incident response in case of a compromise in the ML world...

To shape the future of building tamper-proof ML, join the Coalition for Secure AI, where we are planning to work on building the entire trust ecosystem together with the open source community. In collaboration with multiple industry partners, we are starting up a special interest group under CoSAI for defining the future of ML signing and including tamper-proof ML metadata, such as model cards and evaluation results.

An anonymous reader quotes a report from TechCrunch: In a world filled with "vibe coding," Zach Yadegari, teen founder of Cal AI, stands in ironic, old-fashioned contrast. Ironic because Yadegari and his co-founder, Henry Langmack, are both just 18 years old and still in high school. Yet their story, so far, is a classic. Launched in May, Cal AI has generated over 5 million downloads in eight months, Yadegari says. Better still, he tells TechCrunch that the customer retention rate is over 30% and that the app generated over $2 million in revenue last month. [...]

The concept is simple: Take a picture of the food you are about to consume, and let the app log calories and macros for you. It's not a unique idea. For instance, the big dog in calorie counting, MyFitnessPal, has its Meal Scan feature. Then there are apps like SnapCalorie, which was released in 2023 and created by the founder of Google Lens. Cal AI's advantage, perhaps, is that it was built wholly in the age of large image models. It uses models from Anthropic and OpenAI and RAG to improve accuracy and is trained on open source food calorie and image databases from sites like GitHub.

"We have found that different models are better with different foods," Yadegari tells TechCrunch. Along the way, the founders coded through technical problems like recognizing ingredients from food packages or in jumbled bowls. The result is an app that the creators say is 90% accurate, which appears to be good enough for many dieters. The report says Yadegari began mastering Python and C# in middle school and went on to build his first business in ninth grade -- a website called Totally Science that gave students access to unblocked games (cleverly named to evade school filters). He sold the company at age 16 to FreezeNova for $100,000.

Following the sale, Yadegari immersed himself in the startup scene, watching Y Combinator videos and networking on X, where he met co-founder Blake Anderson, known for creating ChatGPT-powered apps like RizzGPT. Together, they launched Cal AI and moved to a hacker house in San Francisco to develop their prototype.

Camera manufacturers continue to use different proprietary RAW file formats despite the 20-year existence of Adobe's open-source DNG (Digital Negative) format, creating ongoing compatibility challenges for photographers and software developers.

Major manufacturers including Sony, Canon, and Panasonic defended their proprietary formats as necessary for maintaining control over image processing. Sony's product team told The Verge their ARW format allows them "to maximize performance based on device characteristics such as the image sensor and image processing engine." Canon similarly claims proprietary formats enable "optimum processing during image development."

The Verge argues that this fragmentation forces editing software to specifically support each manufacturer's format and every new camera model -- creating delays for early adopters when new cameras launch. Each new device requires "measuring sensor characteristics such as color and noise," said Adobe's Eric Chan.

For what it's worth, smaller manufacturers like Ricoh, Leica, and Sigma have adopted DNG, which streamlines workflow by containing metadata directly within a single file rather than requiring separate XMP sidecar files.

Coreboot 25.03 has been released with support for 22 new motherboards and several other significant updates, including enhanced display handling, USB debugging, RISC-V support, and RAM initialization for older Intel platforms. Phoronix reports: Coreboot 25.03 delivers display handling improvements, a better USB debugging experience, CPU topology updates, various improvements to the open-source RAM initialization for aging Intel Haswell platforms, improved USB Type-C and Thunderbolt handling, various embedded controller (EC) improvements, better RISC-V architecture support, DDR5-7500 support, and many bug fixes across the sprawling Coreboot codebase. More details, including a full list of the supported boards, can be found here.

An anonymous reader quotes a report from Techdirt: Walled Culture has been following closely Italy's poorly designed Piracy Shield system. Back in December we reported how copyright companies used their access to the Piracy Shield system to order Italian Internet service providers (ISPs) to block access to all of Google Drive for the entire country, and how malicious actors could similarly use that unchecked power to shut down critical national infrastructure. Since then, the Computer & Communications Industry Association (CCIA), an international, not-for-profit association representing computer, communications, and Internet industry firms, has added its voice to the chorus of disapproval. In a letter (PDF) to the European Commission, it warned about the dangers of the Piracy Shield system to the EU economy [...]. It also raised an important new issue: the fact that Italy brought in this extreme legislation without notifying the European Commission under the so-called "TRIS" procedure, which allows others to comment on possible problems [...].

As well as Italy's failure to notify the Commission about its new legislation in advance, the CCIA believes that: this anti-piracy mechanism is in breach of several other EU laws. That includes the Open Internet Regulation which prohibits ISPs to block or slow internet traffic unless required by a legal order. The block subsequent to the Piracy Shield also contradicts the Digital Services Act (DSA) in several aspects, notably Article 9 requiring certain elements to be included in the orders to act against illegal content. More broadly, the Piracy Shield is not aligned with the Charter of Fundamental Rights nor the Treaty on the Functioning of the EU -- as it hinders freedom of expression, freedom to provide internet services, the principle of proportionality, and the right to an effective remedy and a fair trial.

Far from taking these criticisms to heart, or acknowledging that Piracy Shield has failed to convert people to paying subscribers, the Italian government has decided to double down, and to make Piracy Shield even worse. Massimiliano Capitanio, Commissioner at AGCOM, the Italian Authority for Communications Guarantees, explained on LinkedIn how Piracy Shield was being extended in far-reaching ways (translation by Google Translate, original in Italian). [...] That is, Piracy Shield will apply to live content far beyond sports events, its original justification, and to streaming services. Even DNS and VPN providers will be required to block sites, a serious technical interference in the way the Internet operates, and a threat to people's privacy. Search engines, too, will be forced to de-index material. The only minor concession to ISPs is to unblock domain names and IP addresses that are no longer allegedly being used to disseminate unauthorized material. There are, of course, no concessions to ordinary Internet users affected by Piracy Shield blunders. In the future, Italy's Piracy Shield will add:
- 30-minute blackout orders not only for pirate sports events, but also for other live content;
- the extension of blackout orders to VPNs and public DNS providers;
- the obligation for search engines to de-index pirate sites;
- the procedures for unblocking domain names and IP addresses obscured by Piracy Shield that are no longer used to spread pirate content;
- the new procedure to combat piracy on the #linear and "on demand" television, for example to protect the #film and #serietv.

Despite promises of more efficient streaming, the AV1 video codec hasn't achieved widespread adoption seven years after its 2018 debut, even with backing from tech giants Netflix, Microsoft, Google, Amazon, and Meta. The Alliance for Open Media (AOMedia) claims AV1 is 30% more efficient than standards like HEVC, delivering higher-quality video at lower bandwidth while remaining royalty-free.

Major services including YouTube, Netflix, and Amazon Prime Video have embraced the technology, with Netflix encoding approximately 95% of its content using AV1. However, adoption faces significant hurdles. Many streaming platforms including Max, Peacock, and Paramount Plus haven't implemented AV1, partly due to hardware limitations. Devices require specific decoders to properly support AV1, though recent products from Apple, Nvidia, AMD, and Intel have begun including them. "In order to get its best features, you have to accept a much higher encoding complexity," Larry Pearlstein, associate professor at the College of New Jersey, told The Verge. "But there is also higher decoding complexity, and that is on the consumer end."

Amazon and United Launch Alliance will launch 27 full-scale satellites on April 9 as part of Amazon's Project Kuiper, marking the company's first major step toward building a global satellite internet network to rival SpaceX's Starlink. GeekWire reports: ULA said the three-hour window for the Atlas V rocket's liftoff from Cape Canaveral Space Force Station's Space Launch Complex 41 in Florida is scheduled to open at noon ET (9 a.m. PT) that day. ULA is planning a live stream of launch coverage via its website starting about 20 minutes ahead of liftoff. Amazon said next week's mission -- known as Kuiper-1 or KA-1 (for Kuiper Atlas 1) -- will put 27 Kuiper satellites into orbit at an altitude of 280 miles (450 kilometers).

ULA launched two prototype Kuiper satellites into orbit for testing in October 2023, but KA-1 will mark Amazon's first full-scale launch of a batch of operational satellites designed to bring high-speed internet access to millions of people around the world. [...] According to Amazon, the Kuiper satellite design has gone through significant upgrades since the prototypes were launched in 2023. Amazon's primary manufacturing facility is in Kirkland, Wash., with some of the components produced at Project Kuiper's headquarters in nearby Redmond.

The mission profile for KA-1 calls for deploying the satellites safely in orbit and establishing ground-to-space contact. The satellites would then use their electric propulsion systems to settle into their assigned orbits at an altitude of 392 miles (630 kilometers), under the management of Project Kuiper's mission operations team in Redmond. Under the current terms of its license from the Federal Communications Commission, Amazon is due to launch 3,232 Kuiper satellites by 2029, with half of those satellites going into orbit by mid-2026.

An anonymous reader quotes a report from 404 Media: A "vibe coded" AI app developed by entrepreneur and Y Combinator group partner Tom Blomfield has generated recipes that gave users instruction on how to make "Cyanide Ice Cream," "Thick White Cum Soup," and "Uranium Bomb," using those actual substances as ingredients. Vibe coding, in case you are unfamiliar, is the new practice where people, some with limited coding experience, rapidly develop software with AI assisted coding tools without overthinking how efficient the code is as long as it's functional. This is how Blomfield said he made RecipeNinja.AI. [...] The recipe for Cyanide Ice Cream was still live on RecipeNinja.AI at the time of writing, as are recipes for Platypus Milk Cream Soup, Werewolf Cream Glazing, Cholera-Inspired Chocolate Cake, and other nonsense. Other recipes for things people shouldn't eat have been removed.

It also appears that Blomfield has introduced content moderation since users discovered they could generate dangerous or extremely stupid recipes. I wasn't able to generate recipes for asbestos cake, bullet tacos, or glue pizza. I was able to generate a recipe for "very dry tacos," which looks not very good but not dangerous. In a March 20 blog on his personal site, Blomfield explained that he's a startup founder turned investor, and while he has experience with PHP and Ruby on Rails, he has not written a line of code professionally since 2015. "In my day job at Y Combinator, I'm around founders who are building amazing stuff with AI every day and I kept hearing about the advances in tools like Lovable, Cursor and Windsurf," he wrote, referring to AI-assisted coding tools. "I love building stuff and I've always got a list of little apps I want to build if I had more free time."

After playing around with them, he wrote, he decided to build RecipeNinja.AI, which can take a prompt as simple as "Lasagna," and generate an image of the finished dish along with a step-by-stape recipe which can use ElevenLabs's AI generated voice to narrate the instruction so the user doesn't have to interact with a device with his tomato sauce-covered fingers. "I was pretty astonished that Windsurf managed to integrate both the OpenAI and Elevenlabs APIs without me doing very much at all," Blomfield wrote. "After we had a couple of problems with the open AI Ruby library, it quickly fell back to a raw ruby HTTP client implementation, but I honestly didn't care. As long as it worked, I didn't really mind if it used 20 lines of code or two lines of code." Having some kind of voice controlled recipe app sounds like a pretty good idea to me, and it's impressive that Blomfield was able to get something up and running so fast given his limited coding experience. But the problem is that he also allowed users to generate their own recipes with seemingly very few guardrails on what kind of recipes are and are not allowed, and that the site kept those results and showed them to other users.

Robotics and machine learning engineer Naveen Kul developed WattWise, a lightweight open-source CLI tool that monitors power usage via smart plugs and throttles system performance based on electricity pricing and peak hours. Tom's Hardware reports: The simple program, called WattWise, came about when Naveen built a dual-socket EPYC workstation with plans to add four GPUs. It's a power-intensive setup, so he wanted a way to monitor its power consumption using a Kasa smart plug. The enthusiast has released the monitoring portion of the project to the public now, but the portion that manages clocks and power will be released later. Unfortunately, the Kasa Smart app and the Home Assistant dashboard was inconvenient and couldn't do everything he desired. He already had a terminal window running monitoring tools like htop, nvtop, and nload, and decided to take matters into his own hands rather than dealing with yet another app.

Naveen built a terminal-based UI that shows power consumption data through Home Assistant and the TP-Link integration. The app monitors real-time power use, showing wattage and current, as well as providing historical consumption charts. More importantly, it is designed to automatically throttle CPU and GPU performance. Naveen's power provider uses Time-of-Use (ToU) pricing, so using a lot of power during peak hours can cost significantly more. The workstation can draw as much as 1400 watts at full load, but by reducing the CPU frequency from 3.7 GHz to 1.5 GHz, he's able to reduce consumption by about 225 watts. (No mention is made of GPU throttling, which could potentially allow for even higher power savings with a quad-GPU setup.)

Results will vary based on the hardware being used, naturally, and servers can pull far more power than a typical desktop -- even one designed and used for gaming. WattWise optimizes the system's clock speed based on the current system load, power consumption as reported by the smart plug, and the time -- with the latter factoring in peak pricing. From there, it uses a Proportional-Integral (PI) controller to manage the power and adapts system parameters based on the three variables. A blog post with more information is available here.

WattWise is also available on GitHub.

Circle, the issuer of the USDC stablecoin, has filed for an IPO aiming for a $5 billion valuation. It marks the company's second attempt at going public amid renewed momentum in the crypto sector and signs of recovery in tech IPO markets. CNBC reports: A prior merger with a special purpose acquisition company (SPAC) collapsed in late 2022 amid regulatory challenges. Since then, Circle has made strategic moves to position itself closer to the heart of global finance, including the announcement last year that it would relocate its headquarters from Boston to One World Trade Center in New York.

Circle reported $1.68 billion in revenue and reserve income in 2024, up from $1.45 billion in 2023 and $772 million in 2022. The company reported net income last year of about $156 million., down from $268 million a year earlier. A successful IPO would make Circle one of the most prominent pure-play crypto companies to list on a U.S. exchange. Coinbase went public through a direct listing in 2021 and has a market cap of about $44 billion.

Mozilla plans to introduce a suite of paid professional services for its open-source Thunderbird email client, transforming the application into a comprehensive communication platform. Dubbed "Thunderbird Pro," the package aims to compete with established ecosystems like Gmail and Office 365 while maintaining Mozilla's commitment to open-source software.

The Pro tier will include four core services: Thunderbird Appointment for streamlined scheduling, Thunderbird Send for file sharing (reviving the discontinued Firefox Send), Thunderbird Assist offering AI capabilities powered by Flower AI, and Thundermail, a revamped email client built on Stalwart's open-source stack. Initially, Thunderbird Pro will be available free to "consistent community contributors," with paid access for other users.

Mozilla Managing Director Ryan Sipes indicated the company may consider limited free tiers once the service establishes a sustainable user base. This initiative follows Mozilla's 2023 announcement about "remaking" Thunderbird's architecture to modernize its aging codebase, addressing user losses to more feature-rich competitors.

An anonymous reader quotes a report from Ars Technica: What does it take to get OpenAI and Anthropic -- two competitors in the AI assistant market -- to get along? Despite a fundamental difference in direction that led Anthropic's founders to quit OpenAI in 2020 and later create the Claude AI assistant, a shared technical hurdle has now brought them together: How to easily connect their AI models to external data sources. The solution comes from Anthropic, which developed and released an open specification called Model Context Protocol (MCP) in November 2024. MCP establishes a royalty-free protocol that allows AI models to connect with outside data sources and services without requiring unique integrations for each service.

"Think of MCP as a USB-C port for AI applications," wrote Anthropic in MCP's documentation. The analogy is imperfect, but it represents the idea that, similar to how USB-C unified various cables and ports (with admittedly a debatable level of success), MCP aims to standardize how AI models connect to the infoscape around them. So far, MCP has also garnered interest from multiple tech companies in a rare show of cross-platform collaboration. For example, Microsoft has integrated MCP into its Azure OpenAI service, and as we mentioned above, Anthropic competitor OpenAI is on board. Last week, OpenAI acknowledged MCP in its Agents API documentation, with vocal support from the boss upstairs. "People love MCP and we are excited to add support across our products," wrote OpenAI CEO Sam Altman on X last Wednesday.

MCP has also rapidly begun to gain community support in recent months. For example, just browsing this list of over 300 open source servers shared on GitHub reveals growing interest in standardizing AI-to-tool connections. The collection spans diverse domains, including database connectors like PostgreSQL, MySQL, and vector databases; development tools that integrate with Git repositories and code editors; file system access for various storage platforms; knowledge retrieval systems for documents and websites; and specialized tools for finance, health care, and creative applications. Other notable examples include servers that connect AI models to home automation systems, real-time weather data, e-commerce platforms, and music streaming services. Some implementations allow AI assistants to interact with gaming engines, 3D modeling software, and IoT devices.

Software engineer Sergey Tselovalnikov weighs in on the new hype: The term caught on and Twitter quickly flooded with posts about how AI has radically transformed coding and will soon replace all software engineers. While AI undeniably impacts the way we write code, it hasn't fundamentally changed our role as engineers. Allow me to explain.

[...] Vibe coding is interacting with the codebase via prompts. As the implementation is hidden from the "vibe coder", all the engineering concerns will inevitably get ignored. Many of the concerns are hard to express in a prompt, and many of them are hard to verify by only inspecting the final artifact. Historically, all engineering practices have tried to shift all those concerns left -- to the earlier stages of development when they're cheap to address. Yet with vibe coding, they're shifted very far to the right -- when addressing them is expensive.

The question of whether an AI system can perform the complete engineering cycle and build and evolve software the same way a human can remains open. However, there are no signs of it being able to do so at this point, and if it one day happens, it won't have anything to do with vibe coding -- at least the way it's defined today.

[...] It is possible that there'll be a future where software is built from vibe-coded blocks, but the work of designing software able to evolve and scale doesn't go away. That's not vibe engineering -- that's just engineering, even if the coding part of it will look a bit different.

OpenAI plans to release a new open-weight language model -- its first since GPT-2 -- in the coming months and is seeking community feedback to shape its development. "That's according to a feedback form the company published on its website Monday," reports TechCrunch. "The form, which OpenAI is inviting 'developers, researchers, and [members of] the broader community' to fill out, includes questions like 'What would you like to see in an open-weight model from OpenAI?' and 'What open models have you used in the past?'" From the report: "We're excited to collaborate with developers, researchers, and the broader community to gather inputs and make this model as useful as possible," OpenAI wrote on its website. "If you're interested in joining a feedback session with the OpenAI team, please let us know [in the form] below." OpenAI plans to host developer events to gather feedback and, in the future, demo prototypes of the model. The first will take place in San Francisco within a few weeks, followed by sessions in Europe and Asia-Pacific regions.

OpenAI is facing increasing pressure from rivals such as Chinese AI lab DeepSeek, which have adopted an "open" approach to launching models. In contrast to OpenAI's strategy, these "open" competitors make their models available to the AI community for experimentation and, in some cases, commercialization.

An anonymous reader quotes a report from 404 Media: The creator of an open source genetic database is shutting it down and deleting all of its data because he has come to believe that its existence is dangerous with "a rise in far-right and other authoritarian governments" in the United States and elsewhere. "The largest use case for DTC genetic data was not biomedical research or research in big pharma," Bastian Greshake Tzovaras, the founder of OpenSNP, wrote in a blog post. "Instead, the transformative impact of the data came to fruition among law enforcement agencies, who have put the genealogical properties of genetic data to use."

OpenSNP has collected roughly 7,500 genomes over the last 14 years, primarily by allowing people to voluntarily submit their own genetic information they have downloaded from 23andMe. With the bankruptcy of 23andMe, increased interest in genetic data by law enforcement, and the return of Donald Trump and rise of authoritarian governments worldwide, Greshake Tzovaras told 404 Media he no longer believes it is ethical to run the database. "I've been thinking about it since 23andMe was on the verge of bankruptcy and been really considering it since the U.S. election. It definitely is really bad over there [in the United States]," Greshake Tzovaras told 404 Media. "I am quite relieved to have made the decision and come to a conclusion. It's been weighing on my mind for a long time."

Greshake Tzovaras said that he is proud of the OpenSNP project, but that, in a world where scientific data is being censored and deleted and where the Trump administration has focused on criminalizing immigrants and trans people, he now believes that the most responsible thing to do is to delete the data and shut down the project. "Most people in OpenSNP may not be at particular risk right now, but there are people from vulnerable populations in here as well," Greshake Tzovaras said. "Thinking about gender representation, minorities, sexual orientation -- 23andMe has been working on the whole 'gay gene' thing, it's conceivable that this would at some point in the future become an issue." "Across the globe there is a rise in far-right and other authoritarian governments. While they are cracking down on free and open societies, they are also dedicated to replacing scientific thought and reasoning with pseudoscience across disciplines," Greshake Tzovaras wrote. "The risk/benefit calculus of providing free & open access to individual genetic data in 2025 is very different compared to 14 years ago. And so, sunsetting openSNP -- along with deleting the data stored within it -- feels like it is the most responsible act of stewardship for these data today."

"The interesting thing to me is there are data preservation efforts in the U.S. because the government is deleting scientific data that they don't like. This is approaching that same problem from a different direction," he added. "We need to protect the people in this database. I am supportive of preserving scientific data and knowledge, but the data comes second -- the people come first. We prefer deleting the data."

Learning to code has become sort of become pointless as AI increasingly dominates programming tasks, said Replit founder and chief executive Amjad Masad. "I no longer think you should learn to code," Masad wrote on X.

The statement comes as major tech executives report significant AI inroads into software development. Google CEO Sundar Pichai recently revealed that 25% of new code at the tech giant is AI-generated, though still reviewed by engineers. Furthermore, Anthropic CEO Dario Amodei predicted AI could generate up to 90% of all code within six months.

Masad called this shift a "bittersweet realization" after spending years popularizing coding through open-source work, Codecademy, and Replit -- a platform that now uses AI to help users build apps and websites. Instead of syntax-focused programming skills, Masad recommends learning "how to think, how to break down problems... how to communicate clearly, with humans and with machines."