Agreguesi i feed

A study of more than 12,000 European firms found that AI adoption causally increases labour productivity by 4% on average across the EU, and that it does so without reducing employment in the short run. Researchers from the Bank for International Settlements and the European Investment Bank used an instrumental variable strategy that matched EU firms to comparable US firms by sector, size, investment intensity and other characteristics, then used the AI adoption rates of those US counterparts as a proxy for exogenous AI exposure among European firms. The productivity gains, however, skewed heavily toward medium and large companies. Among large firms, 45% had deployed AI, compared to just 24% of small firms. The study also found that complementary investments mattered enormously: an extra percentage point of spending on workforce training amplified AI's productivity effect by 5.9%, and an extra point on software and data infrastructure added 2.4%.

Read more of this story at Slashdot.

Cleveland.com, the digital arm of Ohio's Plain Dealer newspaper, has removed writing from the workloads of certain reporters and handed that job to what editor Chris Quinn calls an "AI rewrite specialist" who turns reporter-gathered material into article drafts. The reporters on these beats -- covering Lorain, Lake, Geauga, and most recently Medina County -- are assigned entirely to reporting, spending their time on in-person interviews and meeting sources for coffee. Editors review the AI-produced drafts and reporters get the final say before publication. Quinn says the arrangement has effectively freed up an extra workday per week for each reporter. The newsroom adopted this model last year to expand local coverage into counties it could no longer staff with full teams, and Quinn described the setup in a February 14 letter after a college journalism student withdrew from a reporting role over the newsroom's use of AI. Quinn blamed journalism schools for the student's reaction, saying professors have repeatedly told students that AI is bad.

Read more of this story at Slashdot.

Andrew Yang, the former presidential candidate and longtime Universal Basic Income advocate, published a blog post this week warning that AI is about to displace millions of white-collar workers in the U.S. over the next 12 to 18 months, a wave he has taken to calling "the Fuckening." Yang cited a conversation with the CEO of a publicly traded tech company who said the firm is cutting 15% of its workforce now and plans another 20% cut in two years, followed by yet another 20% two years after that. The U.S. currently has about 70 million white-collar workers, and Yang expects that number to fall by 20 to 50% over the next several years. Underemployment among recent college graduates has already hit 52%, and only 30% of graduating seniors have landed a job in their field. Yang's proposed remedy remains the same one he ran on in 2020: Universal Basic Income.

Read more of this story at Slashdot.

Linus Torvalds has told The Register how Linux went from a solo hobby project on a single 386 PC in Helsinki to a genuinely collaborative effort, and the path involved crowdsourced checks, an FTP mirror at MIT, and a licensing decision that opened the floodgates. Torvalds released the first public snapshot, Linux 0.02, on October 5, 1991, on a Finnish FTP server -- about 10,000 lines of code that he had cross-compiled under Minix. He originally wanted to call it "Freax," but his friend Ari Lemmke, who set up the server, named the directory "Linux" instead. Early contributor Theodore Ts'o set up the first North American mirror on his VAXstation at MIT, since the sole 64 kbps link between Finland and the US made downloads painful. That mirror gave developers on this side of the Atlantic their first practical access to the kernel. Another early developer, Dirk Hohndel, recalled that Torvalds initially threw away incoming patches and reimplemented them from scratch -- a habit he eventually dropped because it did not scale. When Torvalds could not afford to upgrade his underpowered 386, developer H. Peter Anvin collected checks from contributors through his university mailbox and wired the funds to Finland, covering the international banking fees himself. Torvalds got a 486DX/2. In 1992, he moved the kernel to the GPL, and the first full distributions appeared in 1992-1993, turning Linux from a kernel into installable systems.

Read more of this story at Slashdot.

An anonymous reader writes: Electric buses are proving unreliable this winter for Vermont's Green Mountain Transit, as it needs to be over 41 degrees for the buses to charge, but due to a battery recall the buses are a fire hazard and can't be charged in a garage. Spokesman for energy workers advocacy group Power the Future Larry Behrens told the Center Square: "Taxpayers were sold an $8 million 'solution' that can't operate in cold weather when the home for these buses is in New England." "We're beyond the point where this looks like incompetence and starts to smell like fraud," Behrens said. "When government rushes money out the door to satisfy green mandates, basic questions about performance, safety, and value for taxpayers are always pushed aside," Behrens said. "Americans deserve to know who approved this purchase and why the red flags were ignored." General manager at Green Mountain Transit (GMT) Clayton Clark told The Center Square that "the federal government provides public transit agencies with new buses through a competitive grant application process, and success is not a given."

Read more of this story at Slashdot.

Microsoft says a Microsoft 365 Copilot bug has been causing the AI assistant to summarize confidential emails since late January, bypassing data loss prevention (DLP) policies that organizations rely on to protect sensitive information. From a report: According to a service alert seen by BleepingComputer, this bug (tracked under CW1226324 and first detected on January 21) affects the Copilot "work tab" chat feature, which incorrectly reads and summarizes emails stored in users' Sent Items and Drafts folders, including messages that carry confidentiality labels explicitly designed to restrict access by automated tools. Copilot Chat (short for Microsoft 365 Copilot Chat) is the company's AI-powered, content-aware chat that lets users interact with AI agents. Microsoft began rolling out Copilot Chat to Word, Excel, PowerPoint, Outlook, and OneNote for paying Microsoft 365 business customers in September 2025.

Read more of this story at Slashdot.

For a few days leading up to FOSDEM 2026, the GNOME OS developers met for a GNOME OS hackfest. Here are some of the things we talked about!

Stable

The first big topic on our to-do list was GNOME OS stable. We started by defining the milestone: we can call GNOME OS “stable” when we settle on a configuration that we’re willing to support long-term. The most important blocker here is systemd-homed: we know that we want the stable release of GNOME OS to use systemd-homed, and we don’t want to have to support pre-homed GNOME OS installations forever. We discussed the possiblity of building a migration script to move people onto systemd-homed once it’s ready, but it’s simply too difficult and dangerous to deploy this in practice.

We did, however, agree that we can already start promoting GNOME OS a bit more heavily, provided that we make very clear that this is an unstsable product for very early adopters, who would be willing to occasionally reinstall their system (or manually migrate it).

We also discussed the importance of project documentation. GNOME OS’s documentation isn’t in a great state at the moment, and this makes it especially difficult to start contributing. BuildStream, which is GNOME OS’s build system, has a workflow that is unfamiliar to most people that may want to contribute. Despite its comprehensive documentation, there’s no easy “quick start” reference for the most common tasks and so it is ultimately a source of friction for potential contributors. This is especially unfourtunate given the current excitement around building next-gen “distroless” operating systems. Our user documentation is also pretty sparse. Finally, the little documentation we do have is spread across different places (markdown comitted to git, GitLab Wiki pages, the GNOME OS website, etc) and this makes it very difficult for people to find it.

Fixing /etc

Next we talked about the situation with /etc on GNOME OS. /etc has been a bit of an unsolved problem in the UAPI group’s model of immutability: ideally all default configuration can be loaded from /usr, and so /etc would remain entirely for overrides by the system administrator. Unfourtunately, this isn’t currently the case, so we must have some solution to keep track of both upstream defaults and local changes in /etc.

So far, GNOME OS had a complicated set-up where parts of /usr would be symlinked into /etc. To change any of these files, the user would have to break the symlinks and replace them with normal files, potentially requiring copies of entire directories. This would then cause loads of issues, where the broken symlinks cause /etc to slowly drift away from the changing defaults in /usr.

For years, we’ve known that the solution would be overlayfs. This kernel filesystem allows us to mount the OS’s defaults underneath a writable layer for administrator overrides. For various reasons, however, we’ve struggled to deploy this in practice.

Modern systemd has native support for this arrangement via systemd-confext, and we decided to just give it a try at the hackfest. A few hours later, Valentin had a merge request to transition us to the new scheme. We’ve now fully rolled this out, and so the issue is solved in the latest GNOME OS nightlies.

FEX and Flatpak

Next, we discussed integrating FEX with Flatpak so that we can run x86 apps on ARM64 devices.

Abderrahim kicked off the topic by telling us about fexwrap, a script that grafts two different Flatpak runtimes together to successfully run apps via FEX. After studying this implementation, we discussed what proper upstream support might look like.

Ultimately, we decided that the first step will be a new Flatpak runtime extension that bundles FEX, the required extra libraries, and the “thunks” (glue libraries that let x86 apps call into native ARM GPU drivers). From there, we’ll have to experiment and see what integrations Flatpak itself needs to make everything work seamlessly.

Abderrahim has already started hacking on this upstream.

Amutable

The Amutable crew were in Brussels for FOSDEM, and a few of them stopped in to attend our hackfest. We had some very interesting conversations! From a GNOME OS perspective, we’re quite excited about the potential overlap between our work and theirs.

We also used the opportunity to discuss GNOME OS, of course! For instance, we were able to resolve some kernel VFS blockers for GNOME OS delta updates and Flatpak v2.

mkosi

For a few years, we’ve been exploring ways to factor out GNOME OS’s image build scripts into a reusable component. This would make it trivial for other BuildStream-based projects to distribute themselves as UAPI.3 DDIs. It would also allow us to ship device-specific builds of GNOME OS, which are necessary to target mobile devices like the Fairphone 5.

At Boiling the Ocean 7, we decided to try an alternative approach. What if we could drop our bespoke image build steps, and just use mkosi? There, we threw together a prototype and successfully booted to login. With the concept proven, I put together a better prototype in the intervening months. This prompted a discussion with Daan, the maintainer of mkosi, and we ultimately decided that mkosi should just have native BuildStream support upstream.

At the hackfest, Daan put together a prototype for this native support. We were able to use his modified build of mkosi to build a freedesktop-sdk BuildStream image, package it up as a DDI, boot it in a virtual machine, set the machine up via systemd-firstboot, and log into a shell. Daan has since opened a pull request, and we’ll continue iterating on this approach in the coming months.

Overall, this hackfest was extremely productive! I think it’s pretty likely that we’ll organize something like this again next year!

Today, a very quick note on dynamic instance type checks in virtual machines with single inheritance.

The problem is that given an object o whose type is t, you want to check if o actually is of some more specific type u. To my knowledge, there are two sensible ways to implement these type checks.

if the set of types is fixed: dfs numbering

Consider a set of types T := {t, u, ...} and a set of edges S := {<t|ε, u>, ...} indicating that t is the direct supertype of u, or ε if u is a top type. S should not contain cycles and is thus a direct acyclic graph rooted at ε.

First, compute a pre-order and post-order numbering for each t in the graph by doing a depth-first search over S from ε. Something like this:

def visit(t, counter): t.pre_order = counter counter = counter + 1 for u in S[t]: counter = visit(u, counter) t.post_order = counter return counter

Then at run-time, when making an object of type t, you arrange to store the type’s pre-order number (its tag) in the object itself. To test if the object is of type u, you extract the tag from the object and check if tag–u.pre_order mod 2n < u.post_order–u.pre_order.

Two notes, probably obvious but anyway: one, you know the numbering for u at compile-time and so can embed those variables as immediates. Also, if the type has no subtypes, it can be a simple equality check.

Note that this approach applies only if the set of types T is fixed. This is the case when statically compiling a WebAssembly module in a system that doesn’t allow modules to be instantiated at run-time, like Wastrel. Interestingly, it can also be the case in JIT compilers, when modeling types inside the optimizer.

if the set of types is unbounded: the display hack

If types may be added to a system at run-time, maintaining a sorted set of type tags may be too much to ask. In that case, the standard solution is something I learned of as the display hack, but whose name is apparently ungooglable. It is described in a 4-page technical note by Norman H. Cohen, from 1991: Type-Extension Type Tests Can Be Performed In Constant Time.

The basic idea is that each type t should have an associated sorted array of supertypes, starting with its top type and ending with t itself. Each t also has a depth, indicating the number of edges between it and its top type. A type u is a subtype of t if u[t.depth]=t, if u.depth <= t.depth.

There are some tricks one can do to optimize out the depth check, but it’s probably a wash given the check performs a memory access or two on the way. But the essence of the whole thing is in Cohen’s paper; go take a look!

Jan Vitek notes in a followup paper (Efficient Type Inclusion Tests) that Christian Queinnec discovered the technique around the same time. Vitek also mentions the DFS technique, but as prior art, apparently already deployed in DEC Modula-3 systems. The term “display” was bouncing around in the 80s to describe some uses of arrays; I learned it from Dybvig’s implementation of flat closures, who learned it from Cardelli. I don’t know though where “display hack” comes from.

That’s it! If you know of any other standard techniques for type checks with single-inheritance subtyping, do let me know in the comments. Until next time, happy hacking!

Version:next-20260218 (linux-next) Released:2026-02-18