The Trump administration introduced a new framework to expedite self-driving vehicle deployment by reducing regulatory hurdles, while maintaining mandatory safety incident reporting. NHTSA is also expanding its exemption program, allowing domestically produced autonomous vehicles lacking traditional safety controls to operate on U.S. roads. Reuters reports: The Trump administration said Thursday it aims to speed up the deployment of self-driving vehicles but will maintain rules requiring reporting of safety incidents involving advanced vehicles. U.S. Transportation Secretary Sean Duffy on Thursday released a new framework to boost autonomous vehicles. "This administration understands that we're in a race with China to out-innovate, and the stakes couldn't be higher," Duffy said. "Our new framework will slash red tape."

The National Highway Traffic Safety Administration said it will expand a program to exempt some self-driving vehicles from all safety requirements and will streamline but continue its requirement that vehicles equipped with certain advanced driver assistance systems or self-driving systems report safety incidents. NHTSA is expanding its Automated Vehicle Exemption Program to now include domestically produced vehicles that will allow companies to operate non-compliant imported vehicles on U.S. roads. It is currently only open to foreign assembled models.

Microsoft has introduced BitNet b1.58 2B4T, the largest-scale 1-bit AI model to date with 2 billion parameters and the ability to run efficiently on CPUs. It's openly available under an MIT license. TechCrunch reports: The Microsoft researchers say that BitNet b1.58 2B4T is the first bitnet with 2 billion parameters, "parameters" being largely synonymous with "weights." Trained on a dataset of 4 trillion tokens -- equivalent to about 33 million books, by one estimate -- BitNet b1.58 2B4T outperforms traditional models of similar sizes, the researchers claim.

BitNet b1.58 2B4T doesn't sweep the floor with rival 2 billion-parameter models, to be clear, but it seemingly holds its own. According to the researchers' testing, the model surpasses Meta's Llama 3.2 1B, Google's Gemma 3 1B, and Alibaba's Qwen 2.5 1.5B on benchmarks including GSM8K (a collection of grade-school-level math problems) and PIQA (which tests physical commonsense reasoning skills). Perhaps more impressively, BitNet b1.58 2B4T is speedier than other models of its size -- in some cases, twice the speed -- while using a fraction of the memory.

There is a catch, however. Achieving that performance requires using Microsoft's custom framework, bitnet.cpp, which only works with certain hardware at the moment. Absent from the list of supported chips are GPUs, which dominate the AI infrastructure landscape.

Razer's upcoming Blade 16 and other laptops are no longer available for preorder or purchase on its US site. From a report: The configurator for preordering its new Blade 16 laptop was available as recently as April 1st, according to the Internet Archive -- one day before the Trump administration announced sweeping US tariffs on China, Taiwan, and others that make laptop components.

When asked recently if tariffs might affect Razer's prices or availability, its Public Relations Manager, Andy Johnston, told The Verge, "We do not have a comment at this stage regarding tariffs." Razer may not be openly talking about the impact of tariffs, but Framework halted sales of its entry-level Laptop 13 in the US on April 7th, and Micron reportedly confirmed surcharges for its memory chips will apply once the tariffs take effect after midnight tonight.

Framework -- a company that makes upgradeable and repairable laptops -- will pause sales on several versions of one of its models in America thanks to Trump's tariffs, it said. From a report: "Due to the new tariffs that came into effect on April 5th, we're temporarily pausing US sales on a few base Framework Laptop 13 systems (Ultra 5 125H and Ryzen 5 7640U). For now, these models will be removed from our US site. We will continue to provide updates as we have them," Framework said in a post on X.

A spokesperson for Framework told 404 Media in an email that the company was pausing sales on their six lowest priced units in the U.S. They clarified that those models are still available to customers that are ordering the machines outside of America.

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.

France's competition authority has fined Apple 150 million euros ($162 million) for abusing its market dominance through its App Tracking Transparency system, ruling the privacy initiative unfairly disadvantages app developers. The watchdog determined that requiring third-party developers to use two pop-ups for tracking permissions while Apple's own apps need just one tap creates an "excessively complex" process that particularly harms smaller publishers lacking sufficient proprietary data for alternative targeting.

The authority acknowledged the system's privacy benefits, but concluded the framework is "neither necessary nor proportionate" with data protection goals. The regulator is not requiring Apple to modify the system, only imposing the fine for past practices. Apple must display a summary of the decision on its website for seven days.

An anonymous reader quotes a report from Android Authority: No matter the manufacturer, every Android phone has one thing in common: its software base. Manufacturers can heavily customize the look and feel of the Android OS they ship on their Android devices, but under the hood, the core system functionality is derived from the same open-source foundation: the Android Open Source Project. After over 16 years, Google is making big changes to how it develops the open source version of Android in an effort to streamline its development. [...] Beginning next week, all Android development will occur within Google's internal branches, and the source code for changes will only be released when Google publishes a new branch containing those changes. As this is already the practice for most Android component changes, Google is simply consolidating its development efforts into a single branch.

This change will have minimal impact on regular users. While it streamlines Android OS development for Google, potentially affecting the speed of new version development and bug reduction, the overall effect will likely be imperceptible. Therefore, don't expect this change to accelerate OS updates for your phone. This change will also have minimal impact on most developers. App developers are unaffected, as it pertains only to platform development. Platform developers, including those who build custom ROMs, will largely also see little change, since they typically base their work on specific tags or release branches, not the main AOSP branch. Similarly, companies that release forked AOSP products rarely use the main AOSP branch due to its inherent instability.

External developers who enjoy reading or contributing to AOSP will likely be dismayed by this news, as it reduces their insight into Google's development efforts. Without a GMS license, contributing to Android OS development becomes more challenging, as the available code will consistently lag behind by weeks or months. This news will also make it more challenging for some developers to keep up with new Android platform changes, as they'll no longer be able to track changes in AOSP. For reporters, this change means less access to potentially revealing information, as AOSP patches often provide insights into Google's development plans. [...] Google will share more details about this change when it announces it later this week. If you're interested in learning more, be sure to keep an eye out for the announcement and new documentation on source.android.com. Android Authority's Mishaal Rahman says Google is "committed to publishing Android's source code, so this change doesn't mean that Android is becoming closed-source."

"What will change is the frequency of public source code releases for specific Android components," says Rahman. "Some components like the build system, update engine, Bluetooth stack, Virtualization framework, and SELinux configuration are currently AOSP-first, meaning they're developed fully in public. Most Android components like the core OS framework are primarily developed internally, although some features, such as the unlocked-only storage area API, are still developed within AOSP."

An anonymous reader quotes a report from Ars Technica: Anthropic CEO Dario Amodei raised a few eyebrows on Monday after suggesting that advanced AI models might someday be provided with the ability to push a "button" to quit tasks they might find unpleasant. Amodei made the provocative remarks during an interview at the Council on Foreign Relations, acknowledging that the idea "sounds crazy."

"So this is -- this is another one of those topics that's going to make me sound completely insane," Amodei said during the interview. "I think we should at least consider the question of, if we are building these systems and they do all kinds of things like humans as well as humans, and seem to have a lot of the same cognitive capacities, if it quacks like a duck and it walks like a duck, maybe it's a duck."

Amodei's comments came in response to an audience question from data scientist Carmem Domingues about Anthropic's late-2024 hiring of AI welfare researcher Kyle Fish "to look at, you know, sentience or lack of thereof of future AI models, and whether they might deserve moral consideration and protections in the future." Fish currently investigates the highly contentious topic of whether AI models could possess sentience or otherwise merit moral consideration. "So, something we're thinking about starting to deploy is, you know, when we deploy our models in their deployment environments, just giving the model a button that says, 'I quit this job,' that the model can press, right?" Amodei said. "It's just some kind of very basic, you know, preference framework, where you say if, hypothesizing the model did have experience and that it hated the job enough, giving it the ability to press the button, 'I quit this job.' If you find the models pressing this button a lot for things that are really unpleasant, you know, maybe you should -- it doesn't mean you're convinced -- but maybe you should pay some attention to it."

Amodei's comments drew immediate skepticism on X and Reddit.

An anonymous reader quotes a report from Ars Technica: On Tuesday, OpenAI unveiled a new "Responses API" designed to help software developers create AI agents that can perform tasks independently using the company's AI models. The Responses API will eventually replace the current Assistants API, which OpenAI plans to retire in the first half of 2026. With the new offering, users can develop custom AI agents that scan company files with a file search utility that rapidly checks company databases (with OpenAI promising not to train its models on these files) and navigate websites -- similar to functions available through OpenAI's Operator agent, whose underlying Computer-Using Agent (CUA) model developers can also access to enable automation of tasks like data entry and other operations.

However, OpenAI acknowledges that its CUA model is not yet reliable for automating tasks on operating systems and can make unintended mistakes. The company describes the new API as an early iteration that it will continue to improve over time. Developers using the Responses API can access the same models that power ChatGPT Search: GPT-4o search and GPT-4o mini search. These models can browse the web to answer questions and cite sources in their responses. That's notable because OpenAI says the added web search ability dramatically improves the factual accuracy of its AI models. On OpenAI's SimpleQA benchmark, which aims to measure confabulation rate, GPT-4o search scored 90 percent, while GPT-4o mini search achieved 88 percent -- both substantially outperforming the larger GPT-4.5 model without search, which scored 63 percent.

Despite these improvements, the technology still has significant limitations. Aside from issues with CUA properly navigating websites, the improved search capability doesn't completely solve the problem of AI confabulations, with GPT-4o search still making factual mistakes 10 percent of the time. Alongside the Responses API, OpenAI released the open source Agents SDK, providing developers free tools to integrate models with internal systems, implement safeguards, and monitor agent activities. This toolkit follows OpenAI's earlier release of Swarm, a framework for orchestrating multiple agents.

An anonymous reader quotes a report from TechCrunch: When Bluesky CEO Jay Graber walked on stage at SXSW 2025 for her keynote discussion, she wore a large black T-shirt with her hair pulled back into a bun. At first glance, it might appear as though she's following the same playbook that so many women in tech leadership have played before: downplaying her femininity to be taken seriously. The truth is way more interesting than that. What might look like your average black T-shirt is a subtle, yet clear swipe at Mark Zuckerberg, a CEO who represents everything that Bluesky is trying to work against as an open source social network.

The Meta founder and CEO has directly compared himself to the Roman emperor Julius Caesar. His own shirt declared Aut Zuck aut nihil, which is a play on the Latin phrase aut Caesar aut nihil: "Either Caesar or nothing." Graber's shirt -- which directly copies the style of a shirt that Zuckerberg wore onstage recently -- says Mundus sine caesaribus. Or, "a world without Caesars." With the way Bluesky is designed, Graber is certainly putting her money where her mouth (or shirt) is. As a decentralized social network built upon an open source framework, Bluesky differs from legacy platforms like Facebook in that users have a direct, transparent window into how the platform is being built. "If a billionaire came in and bought Bluesky, or took it over, or if I decided tomorrow to change things in a way that people really didn't like, then they could fork off and go on to another application," Graber explained at SXSW. "There's already applications in the network that give you another way to view the network, or you could build a new one as well. And so that openness guarantees that there's always the ability to move to a new alternative."

Today marks the release of Framework Mono 6.14, the first major Mono release in five years and the first under WineHQ's stewardship. This update includes long-awaited improvements such as native macOS ARM64 support, enhancements to System.Windows.Forms for X11, better COM interface support, and various stability fixes. Phoronix reports: In addition to the native macOS ARM64 support and System.Windows.Forms improvements for X11, some of the other Mono 6.14 improvements carried out over the past half-decade include improved support for generated COM interfaces, many warning fixes, addressing common cases where processes would hang on exit, and more.

As for the "Framework Mono" name rather than just Mono, the release announcement explains: "Framework Mono is the project previously hosted at https://github.com/mono/mono, which was then simply called Mono. I have made this change to distinguish it from 'monovm' and 'Wine Mono', which are different projects. Framework Mono is a cross-platform runtime compatible with .NET Framework." You can download and learn more about the release here.

Google has introduced a Debian Linux terminal app for Android in its ongoing effort to transform Android into a versatile desktop OS. It's initially available on Pixel devices running Android 15 but will be expanded to "all sufficiently robust Android phones" when Android 16 arrives later this year," writes ZDNet's Steven Vaughan-Nichols. An anonymous reader shares an excerpt from the report: Today, Linux is only available on the latest Pixel devices running Android 15. When Android 16 arrives later this year, it's expected that all sufficiently robust Android phones will be able to run Linux. Besides a Linux terminal, beta tests have already shown that you should be able to run desktop Linux programs from your phone -- games like Doom, for example. The Linux Terminal runs on top of a Debian Linux virtual machine. This enables you to access a shell interface directly on your Android device. And that just scratches the surface of Google's Linux Terminal. It's actually a do-it-all app that enables you to download, configure, and run Debian. Underneath Terminal runs the Android Virtualization Framework (AVF). These are the APIs that enable Android devices to run other operating systems.

To try the Linux Terminal app, you must activate Developer Mode by navigating to Settings - About Phone and tapping the build number seven times. I guess Google wants to make sure you want to do this. Once Developer Mode is enabled, the app can be activated via Settings - System - Developer options - Linux development environment. The initial setup may take a while because it needs to download Debian. Typically this is a 500MB download. Once in place, it allows you to adjust disk space allocation, set port controls for network communication, and recover the virtual machine's storage partition. However, it currently lacks support for graphical user interface (GUI) applications. For that, we'll need to wait for Android 16.

According to Android specialist Mishaal Rahman, 'Google wants to turn Android into a proper desktop operating system, and in order to do that, it has to make it work better with traditional PC input methods and display options. Therefore, Google is now testing new external display management tools in Android 16 that bring Android closer to other desktop OSes.'

Memory safety bugs are "eroding trust in technology and costing billions," argues a new post on Google's security blog — adding that "traditional approaches, like code auditing, fuzzing, and exploit mitigations — while helpful — haven't been enough to stem the tide."

So the blog post calls for a "common framework" for "defining specific, measurable criteria for achieving different levels of memory safety assurance." The hope is this gives policy makers "the technical foundation to craft effective policy initiatives and incentives promoting memory safety" leading to "a market in which vendors are incentivized to invest in memory safety." ("Customers will be empowered to recognize, demand, and reward safety.")

In January the same Google security researchers helped co-write an article noting there are now strong memory-safety "research technologies" that are sufficiently mature: memory-safe languages (including "safer language subsets like Safe Buffers for C++"), mathematically rigorous formal verification, software compartmentalization, and hardware and software protections. (With hardware protections including things like ARM's Memory Tagging Extension and the (Capability Hardware Enhanced RISC Instructions, or "CHERI", architecture.) Google's security researchers are now calling for "a blueprint for a memory-safe future" — though Importantly, the idea is "defining the desired outcomes rather than locking ourselves into specific technologies."

Their blog post this week again urges a practical/actionable framework that's commonly understood, but one that supports different approaches (and allowing tailoring to specific needs) while enabling objective assessment: At Google, we're not just advocating for standardization and a memory-safe future, we're actively working to build it. We are collaborating with industry and academic partners to develop potential standards, and our joint authorship of the recent CACM call-to-action marks an important first step in this process... This commitment is also reflected in our internal efforts. We are prioritizing memory-safe languages, and have already seen significant reductions in vulnerabilities by adopting languages like Rust in combination with existing, wide-spread usage of Java, Kotlin, and Go where performance constraints permit. We recognize that a complete transition to those languages will take time. That's why we're also investing in techniques to improve the safety of our existing C++ codebase by design, such as deploying hardened libc++.

This effort isn't about picking winners or dictating solutions. It's about creating a level playing field, empowering informed decision-making, and driving a virtuous cycle of security improvement... The journey towards memory safety requires a collective commitment to standardization. We need to build a future where memory safety is not an afterthought but a foundational principle, a future where the next generation inherits a digital world that is secure by design.
The security researchers' post calls for "a collective commitment" to eliminate memory-safety bugs, "anchored on secure-by-design practices..." One of the blog post's subheadings? "Let's build a memory-safe future together."

And they're urging changes "not just for ourselves but for the generations that follow."

Technicolor Group has filed for a court recovery procedure in France after failing to secure new investors, putting its VFX brands, including MPC, The Mill, Mikros Animation, and Technicolor Games, at risk of closure. Variety reports: A total shutdown of MPC and Technicolor's operations would affect thousands of visual effects workers in countries include the U.S., UK, Canada and India. The turn in business has raised the alarm and sparked sadness within the VFX community. Parot's memo explains, "In each country, the appropriate framework for orderly protection and way forward is currently being put in place to allow, when possible, to remain in business continuity."

Technicolor has already started to shut down U.S. operations. On Friday, it began alerting customers and employees, sending U.S. employees a WARN notice as required by law for large companies that anticipate closings and mass layoffs. At least one recovery effort already started for roughly 100 U.S. employees of The Mill. The creative leadership and most of the creative staff that was Technicolor's The Mill U.S. is joining forces with Dream Machine FX to launch a new venture, Arc Creative, Variety reported exclusively on Monday. A statement from the artists explains they they are working to launch the new entity amid "the complexities of Technicolor's Chapter 7 proceedings."

Questions remain about how studios will finish upcoming projects that are currently housed at MPC, which include Disney's live-action remake of "Lilo and Stitch" and Paramount's "Mission: Impossible -- The Final Reckoning," as well as Mikros' work, such as Paramount and Nickelodeon's upcoming "Teenage Mutant Ninja Turtles" sequel.

A German nuclear fusion startup called Proxima Fusion has unveiled its "Stellaris" fusion power plant designed to operate reliably and continuously without the instabilities of tokamaks. It's backed by $65 million in funding, with plans to build a fully operational fusion reactor by 2031. TechCrunch reports: Tokamaks and stellarators are types of fusion reactors that use electromagnets to contain fusion plasma. Tokamaks rely on external magnets and an induced plasma current but are known for instability. Stellarators, by contrast, use only external magnets, which, in theory, enable better stability and continuous operation. However, according to Dr. Francesco Sciortino, co-founder and CEO of Proxima Fusion, Proxima's "Stellaris" design is the first peer-reviewed fusion power plant concept that demonstrates it can operate reliably and continuously, without the instabilities and disruptions seen in tokamaks and other approaches.

Proxima published its findings in Fusion Engineering and Design, choosing to share this information publicly to support open-source science. "Our American friends can see it. Our Chinese friends can see it. Our claim is that we can execute on this faster than anyone else, and we do that by creating a framework for integrated physics, engineering, and economics. So we're not a science project anymore," Sciortino told TechCrunch over a call. "We started out as a group of founders saying it's going to take us two years to get to the Stellaris design ... We actually finished after one year. So we've accelerated by a year," he added.

At its "Second Gen" event today, Framework detailed three new computers: an updated Framework Laptop 13 with AMD Ryzen AI 300, a 4.5-liter Mini-ITX desktop powered by Ryzen AI Max, and a colorful, convertible Framework Laptop 12 designed with students in mind. The latter is defined by Framework as a "defining product." Tom's Hardware reports: Framework Desktop: The Framework Desktop is a 4.5L Mini-ITX machine using AMD's Ryzen AI Max "Strix Halo" chips with Radeon 8060S graphics. While this is a mobile chip, Framework says putting it in a desktop chassis gets it to 120W sustained power and 140W boost "while staying quiet and cool." Framework says this should allow 1440p gaming on intense titles, as well as workstation-class projects and local AI. [...] The base model, with a Ryzen AI Max 385 and 32GB of RAM, starts at $1,099, while the top-end machine with a Ryzen AI Max+ 395 with 128GB of RAM begins at $1,999. Framework is only doing "DIY" editions here, so you'll have to get your own storage drive and bring your own operating system (the company is calling it "the easiest PC you'll ever build"). The mainboard on its own will be available from $799. Pre-orders are open now, and Framework expects to ship sometime in Q3.

Framework Laptop 12: The Laptop 12 is designed to bring the flexibility from the Laptop 12 but make it smaller, cheaper, and in more colors (with an optional stylus to match). These machines are made of ABS plastic molded in thermoplastic polyurethane, all around a metal frame. Framework says that it's "our easiest product ever to repair," but that more information on that will come closer to its launch in mid-2025. I'm really looking forward to this repair guide. It comes in five colorways: lavender, sage, gray, black, and bubblegum. The laptop will come with 13th Gen Intel Core i3 and i5 processors, which aren't the latest, but better than entry-level junk. You'll get up to 48GB of RAM, 2TB of storage, and Wi-Fi 6E. It has a 1920 x 1200 touch screen that the company claims will surpass 400 nits of brightness. There's no pricing information yet, and Framework says there's more to share on pricing and specs later in the year. Pre-orders will open in April ahead of the mid-year launch.

Framework Laptop 13: The Framework Laptop 13 is getting a significant refresh with AMD Ryzen AI 300 Series. It doesn't look all that different on the outside, with a 13.5-inch design that largely resembles the one from way back in 2021. But there are new features. Beyond the processors, the Framework Laptop 3 is getting bumped up to Wi-Fi 7 and is getting a new thermal system, a "next-generation" keyboard, and new colorways for the Expansion Cards and bezels (though I still don't know why you would want a bezel in anything other than black). [...] The new Framework Laptop 13 with AMD Ryzen AI 300 starts at $899 for a DIY Edition without storage or an OS, and $1,099 for a pre-built model. If you're buying the mainboard to put in an old system, that's $449. (Framework is keeping the Ryzen 7040 systems around starting at $749). No word for now on any new Intel models.

An anonymous reader quotes a report from Ars Technica: On Tuesday, Hugging Face researchers released an open source AI research agent called "Open Deep Research," created by an in-house team as a challenge 24 hours after the launch of OpenAI's Deep Research feature, which can autonomously browse the web and create research reports. The project seeks to match Deep Research's performance while making the technology freely available to developers. "While powerful LLMs are now freely available in open-source, OpenAI didn't disclose much about the agentic framework underlying Deep Research," writes Hugging Face on its announcement page. "So we decided to embark on a 24-hour mission to reproduce their results and open-source the needed framework along the way!"

Similar to both OpenAI's Deep Research and Google's implementation of its own "Deep Research" using Gemini (first introduced in December -- before OpenAI), Hugging Face's solution adds an "agent" framework to an existing AI model to allow it to perform multi-step tasks, such as collecting information and building the report as it goes along that it presents to the user at the end. The open source clone is already racking up comparable benchmark results. After only a day's work, Hugging Face's Open Deep Research has reached 55.15 percent accuracy on the General AI Assistants (GAIA) benchmark, which tests an AI model's ability to gather and synthesize information from multiple sources. OpenAI's Deep Research scored 67.36 percent accuracy on the same benchmark with a single-pass response (OpenAI's score went up to 72.57 percent when 64 responses were combined using a consensus mechanism).

As Hugging Face points out in its post, GAIA includes complex multi-step questions such as this one: "Which of the fruits shown in the 2008 painting 'Embroidery from Uzbekistan' were served as part of the October 1949 breakfast menu for the ocean liner that was later used as a floating prop for the film 'The Last Voyage'? Give the items as a comma-separated list, ordering them in clockwise order based on their arrangement in the painting starting from the 12 o'clock position. Use the plural form of each fruit." To correctly answer that type of question, the AI agent must seek out multiple disparate sources and assemble them into a coherent answer. Many of the questions in GAIA represent no easy task, even for a human, so they test agentic AI's mettle quite well. Open Deep Research "builds on OpenAI's large language models (such as GPT-4o) or simulated reasoning models (such as o1 and o3-mini) through an API," notes Ars. "But it can also be adapted to open-weights AI models. The novel part here is the agentic structure that holds it all together and allows an AI language model to autonomously complete a research task."

The code has been made public on GitHub.

The DeepComputing RISC-V Mainboard that Framework announced last year for its 13-inch laptops is now available for $199. Liliputing reports: If you already have a Framework Laptop 13 with an Intel or AMD motherboard, the new board is a drop-in replacement. But if you don't have a Framework Laptop you can also use the mainboard as a standalone computer: Framework sells a $39 Cooler Master case that effectively turns its mainboards into mini desktop computers. The RISC-V Mainboard comes from a partnership between Framework and DeepComputing, the Chinese company behind the DC-ROMA laptops, which were some of the first notebook computers to ship with RISC-V processors.

The board features a StarFive JH7110 processor, which is a 1.5 GHz quad-core chip featuring SiFive U74 RISC-V CPU cores and Imagination BXE-4-32 graphics, 8GB of onboard RAM, and a a 64GB SD card for storage (there's also support for an optional eMMC module, but you'll need to bring your own). Since the board is designed to fit in existing laptop frames, it's the same size and shape as AMD or Intel models and has four USB ports in the same locations. But these ports are a little less versatile than the ones you might find on other Framework Laptop 13 Mainboards [...]. There's also a 3.5mm audio jack. You can check out the new board via the Framework Marketplace.

Further reading: Late last year, Framework CEO Nirav Patel delivered one of the best live demos we've ever seen at a tech conference -- modifying a Framework Laptop from x86 to RISC-V live on stage.

NetChoice has filed (PDF) its 10th lawsuit challenging state internet regulations, this time opposing Maryland's Age-Appropriate Design Code Act. The Verge's Lauren Feiner reports: NetChoice has become one of the fiercest -- and most successful -- opponents of age verification, moderation, and design code laws, all of which would put new obligations on tech platforms and change how users experience the internet. [...] NetChoice's latest suit opposes the Maryland Age-Appropriate Design Code Act, a rule that echoes a California law of a similar name. In the California litigation, NetChoice notched a partial win in the Ninth Circuit Court of Appeals, which upheld the district court's decision to block a part of the law requiring platforms to file reports about their services' impact on kids. (It sent another part of the law back to the lower court for further review.)

A similar provision in Maryland's law is at the center of NetChoice's complaint. The group says that Maryland's reporting requirement lets regulators subjectively determine the "best interests of children," inviting "discriminatory enforcement." The reporting requirement on tech companies essentially mandates them "to disparage their services and opine on far-ranging and ill-defined harms that could purportedly arise from their services' 'design' and use of information," NetChoice alleges. NetChoice points out that both California and Maryland have passed separate online privacy laws, which NetChoice Litigation Center director Chris Marchese says shows that "lawmakers know how to write laws to protect online privacy when what they want to do is protect online privacy."

Supporters of the Maryland law say legislators learned from California's challenges and "optimized" their law to avoid questions about speech, according to Tech Policy Press. In a blog analyzing Maryland's approach, Future of Privacy Forum points out that the state made some significant changes from California's version -- such as avoiding an "express obligationâ to determine users' ages and defining the "best interests of children." The NetChoice challenge will test how well those changes can hold up to First Amendment scrutiny. NetChoice has consistently maintained that even well-intentioned attempts to protect kids online are likely to backfire. Though the Maryland law does not explicitly require the use of specific age verification tools, Marchese says it essentially leaves tech platforms with a no-win decision: collect more data on users to determine their ages and create varied user experiences or cater to the lowest common denominator and self-censor lawful content that might be considered inappropriate for its youngest users. And similar to its arguments in other cases, Marchese worries that collecting more data to identify users as minors could create a "honey pot" of kids' information, creating a different problem in attempting to solve another.

President Trump signed an executive order on Thursday that "sets a federal agenda meant to move U.S. digital assets businesses into friendly oversight," reports CoinDesk. The order creates a cryptocurrency working group tasked with proposing a new regulatory framework for digital assets. It will be "made up of the Treasury secretary, attorney general and chairs of the Securities and Exchange Commission and Commodity Futures Trading Commission, along with other agency heads," notes Reuters.

The directive also explores the creation of a "national digital asset stockpile," orders protections for banking services for crypto companies, and bans the creation of central bank digital currencies which could compete with existing cryptocurrencies.