I noticed this a couple of days ago on Slashdot, but didn't have the time to write about it here. The Wall Street Journal wrote about how the Internet rewired work and the implications for the current AI dilemma regarding jobs. The argument has two parts. First of all, about the impact of the Internet on the labor market:

Networked machines and web-enabled software quietly transformed back offices as much as our on-screen lives. Similarly, as e-commerce took off, internet-enabled logistics rewired planning roles — logisticians, transportation and distribution managers — and unlocked a surge in last-mile work. The build-out didn't just hire coders; it hired coordinators, pickers, packers and drivers. It spawned hundreds of thousands of warehouse and delivery jobs — the largest pockets of internet-driven job growth, and yet few had them on their 1998 bingo card... Today, the share of workers in professional and managerial occupations has more than doubled since the dawn of the digital era.

Then, they consider what the lessons might be for a potentially AI-centric labor market:

So what does that tell us about AI? Our mental model often defaults to an industrial image — John Henry versus the steam drill — where jobs are one dominant task, and automation maps one-to-one: Automate the task, eliminate the job. The internet revealed a different reality: Modern roles are bundles. Technologies typically hit routine tasks first, then workflows, and only later reshape jobs, with second-order hiring around the backbone. That complexity is what made disruption slower and more subtle than anyone predicted. AI fits that pattern more than it breaks it... [LLMs] can draft briefs, summarize medical notes and answer queries. Those are tasks — important ones — but still parts of larger roles. They don't manage risk, hold accountability, reassure anxious clients or integrate messy context across teams. Expect a rebalanced division of labor: The technical layer gets faster and cheaper; the human layer shifts toward supervision, coordination, complex judgment, relationship work and exception handling.

What to expect from AI, then, is messy, uneven reshuffling in stages. Some roles will contract sharply — and those contractions will affect real people. But many occupations will be rewired in quieter ways. Productivity gains will unlock new demand and create work that didn't exist, alongside a build-out around data, safety, compliance and infrastructure.

AI is unprecedented; so was the internet. The real risk is timing: overestimating job losses, underestimating the long, quiet rewiring already under way, and overlooking the jobs created in the backbone. That was the internet's lesson. It's likely to be AI's as well.

Or, to put it a different way, we tend to overestimate how disruptive our present is, mainly because, of course, the past already occurred and the future hasn't happened yet, so the present is the only thing that can truly have an impact. Thus, we tend to forget the lessons from the past, even our most recent past. In this case, as the Wall Street Journal argues, I'd have to agree that the Internet also introduced a significant amount of disruption back in the late 1990s and early 2000s, including its own hype and financial bubble. Yet, although it undoubtedly destroyed some jobs, it also created others and, above all, transformed most. It sounds reasonable to expect that AI will have a similar effect on the labor market. Especially since, as the newspaper argues, LLMs can currently do certain tasks, but always with human supervision and initiative. {link to this entry}

Firefox recently added a new AI window feature that, to be honest, doesn't look too bad to me. Chances are I have no use for it. But other people may. It just adds the capability to open a new window to interact with AI, the same way one can now open a new private window. But, as far as I can tell, some users have been complaining quite vocally about it. I'm not sure I understand them, to be clear. When it comes to AI, I don't agree with the AI-hype side nor the AI-is-evil side. As it tends to happen on most topis, I see myself somewhere in between. In this case, closer to the AI skeptic side than the AI fan side. But still somewhere in the middle. But, in particular, I don't understand why these users are complaining so much when, as far as I can tell, Mozilla is doing things right. Not only are they using local AI but, on top of that, the features are always opt-in. Sometimes, I get the feeling that certain people out there have taken a partisan approach to AI the same way other people are taking an overly partisan approach to politics. It's almost as if being pro or against AI is an intrinsic part of their identity. As Jolie Huang, senior staff product manager at Mozilla, recently explained:

"We recently shared how we're approaching AI in Firefox with user choice and openness at the center of everything we build," Huang wrote. "We've heard from many of you who'd prefer not to have AI in your browser at all, and we get it: We will soon provide additional settings for you to control how AI is used (or not) in Firefox.

"Nonetheless, standing still while technology moves forward doesn't benefit the web or the people who use it. That's why we see it as our responsibility to shape how AI integrates into the web, in ways that promote openness, transparency, and choice. That way, users and developers can use it freely, help shape it, and truly benefit from it."

Let's be clear, folks. AI is here to stay. No, it won't save the world. No, it won't take us right away to a scifi utopia. And yes, it will have plenty of negative side-effects. It's already having them. But it will also be useful forcertain taks. I think it's silly to approach this with a religious mindset. {link to this entry}

Using scp when copying files remotely is so ingrained in me, as well as the idea that it comes with the trusted OpenSSH suite, that I never worried much about it. Yet, today, while reading an article on It's FOSS about deprecated Linux commands, I was quite surprised to learn that there are quite a few issues —even security issues!— involving scp. The gory details can be found in this other article by Linux Weekly News. From the looks of it, it's not something extremely dangerous. But it may be a good idea to get into the habit of using either sftp or even rsync as a replacement. {link to this entry}

Clever. I read today that Nvidia connects AI GPUs to early quantum processors as a way to guarantee it still has a future if quantum computing spreads.

Nvidia's open-source software layer, lets researchers choreograph that link — running AI models, quantum algorithms, and error-correction routines together as one system. That jump allows artificial intelligence to monitor [in real time]... For researchers, that means hundreds of new iterations where there used to be one — a genuine acceleration of discovery. It's the quiet kind of progress engineers love — invisible, but indispensable...

Its GPUs (graphics processing units) are already tuned for the dense, parallel calculations these explorations demand, making them the natural partner for any emerging quantum processor... Other companies chase better quantum hardware — superconducting, photonic, trapped-ion — but all of them need reliable coordination with the computing power we already have. By offering that link, Nvidia turns its GPU ecosystem into the operating environment of hybrid computing, the connective tissue between what exists now and what's coming next. And because the system is open, every new lab or start-up that connects strengthens Nvidia's position as the default hub for quantum experimentation...

There's also a defensive wisdom in this move. If quantum computing ever matures, it could threaten the same data center model that built Nvidia's empire. CEO Jensen Huang seems intent on making sure that, if the future shifts, Nvidia already sits at its center. By owning the bridge between today's technology and tomorrow's, the company ensures it earns relevance — and revenue — no matter which computing model dominates.

If I understood this correctly, instead of betting on a full court migration to quantum computing, Nvidia is betting on a long transition that will be hybrid, which sounds like a very smart move to me. Also, I'm not sure if I'm understanding this other part correctly, but perhaps they are also using their GPUs and their software to aid in the badly needed error correction of quantum chips. Again, a pretty smart move in case the long expected quantum revolution happens in the next decade or so. {link to this entry}

I recently read an article on The Register about how using AI for learning truly doesn't appear to work:

A study of how people use ChatGPT for research has confirmed something most of us learned the hard way in school: to be a subject matter expert, you've got to spend time swotting up.

More than 10,000 participants took part in a series of experiments designed to determine how people's understanding of a subject differed when using ready-made summaries from AI chatbots, versus piecing together online information found through traditional web searches.

It found that participants who used ChatGPT and similar tools developed a shallower grasp of the subject they were assigned to study, could provide fewer concrete facts, and tended to echo information similar to other participants who'd used AI tools.

The researchers concluded that, while large language models (LLMs) are exceptionally good at spitting out fluent answers at the press of a button, people who rely on synthesized AI summaries for research typically don't come away with materially deeper knowledge. Only by digging into sources and piecing information together themselves do people tend to build the kind of lasting understanding that sticks, the team found.

Oh, surprise! To learn, one needs to do some work. Put some effort. Isn't that pretty much traditional wisdom?

Interestingly enough, this is true even when AI is just used to do the research:

"In contrast to web search, when learning from LLM summaries users no longer need to exert the effort of gathering and distilling different informational sources on their own — the LLM does much of this for them," the researchers said in a paper published in October's issue of PNAS Nexus.

"We predict that this lower effort in assembling knowledge from LLM syntheses (vs. web links) risks suppressing the depth of knowledge that users gain, which subsequently affects the nature of the advice they form on the topic for others."

In other words: when you outsource the work of research to generative AI, you bypass the mental effort that turns information-gathering into genuine understanding.

Again, none of that should truly surprise us. Said that, I remain convinced that few things in this life are purely good or bad, I think I may have recently come across an instance where using AI for learning makes sense: to dig deeper into a book. Let me be clear. I'm not talking about using AI to summarize the book for us. Rather, what I'm talking about is asking AI to share some analysis (as opposed to a mere summary) chapter by chapter as we read. I've applied this method while reading Neuromancer and Count Zero recently, and I feel as if it was quite useful. It helped me gain a more in-depth view of the books than with traditional reading, which I had done before with both books. {link to this entry}

Now, some of this is way over my head. Still, it's sort of interesting. Apparently, according to a recently publilshed scientific paper there is now clear mathematical proof that our universe cannot be a simulation. Along the way, we learn a few interesting tidbits of information about the nature of our universe.

The research hinges on a fascinating property of reality itself. Modern physics has moved far beyond Newton's tangible "stuff" bouncing around in space. Einstein's theory of relativity replaced Newtonian mechanics. Quantum mechanics transformed our understanding again. Today's cutting-edge theory—quantum gravity—suggests that even space and time aren't fundamental. They emerge from something deeper: pure information.

This information exists in what physicists call a Platonic realm—a mathematical foundation more real than the physical universe we experience. It's from this realm that space and time themselves emerge.

Here's where it gets interesting. The team demonstrated that even this information-based foundation cannot fully describe reality using computation alone. They used powerful mathematical theorems—including Gödel's incompleteness theorem—to prove that a complete and consistent description of everything requires what they call "non-algorithmic understanding."

Think of it this way. A computer follows recipes, step by step, no matter how complex. But some truths can only be grasped through non-algorithmic understanding—understanding that doesn't follow from any sequence of logical steps. These "Gödelian truths" are real, yet impossible to prove through computation.

In other words, together with a matrix-like world, we should also forget about a purely computable universe. Or at least that's how I read it. I could be wrong, of course. As I said, I don't pretend to be able to understand all of it. But there is one more reason to be suspicious of a purely computable universe, I think. As a matter of fact, there may even be a reason to be suspicious of the idea that information underlies our whole universe, as the authors of the paper appear to accept. Sometimes I wonder if perhaps we're just too immmersed in our own paradigm (used here in the sense used by Thomas S. Kuhn), and due to an inherent bias everything around us looks like information. We life in a digital age, after all. {link to this entry}

I upgraded my laptop to Debian 13 (trixie) over the weekend and, for the most part, everything went fine. The actual upgrade was easy enough, and I didn't run into any issues. Today, as I started using it, I ran into a couple of issues with neovim. First of all, I used to be able to right-click on any URL inside a neovim session on the previous release of Debian, and easily click on "Copy URL". However, that feature appeared to be missing now. The fix for this issue was easy: from now on, simply hold down the Shift key while right-clicking on the URL to see the default context menu from the terminal allowing you to copy the URL. The second issue is merely aesthetic, but a bit annoying. For whatever reason, nvim now overrides the default theme from the terminal, and it uses its own background color. This other issue I managed to fix by adding the following line to my configuration file:

highlight Normal guibg=NONE guifg=NONE ctermbg=NONE ctermfg=NONE

{link to this entry}

I've been using Proton's services for quite a few years now. Almost from the beginning. I've seen their ecosystem grow to its current point, where it's almost a pretty good replacement for the Google suite of applications. Yes, sure, it's not as feature-rich. But I can use them on a daily basis without missing good-old Google, to be honest. I may have less features, and the experience may not be as polished. But, in exchange, I don't have a big fat corporate agent breathing down my neck and stealing all my personal information on a daily basis. In any case, they announced a new cool service this week: the Data Breach Observatory. They are going directly to the dark web to capture and share leaks in real time. That's a pretty cool idea, I think. It's amazing that nobody came up with it before. Here is to hoping that the Proton folks continue their great work and manage to survive in this crazy digital world of ours. {link to this entry}

Tonight, I decided to bite the bullet and upgrade to Debian 13 (trixie). I've been running Debian for about 20 years now. But, in the past, instead of performing an actual upgrade to go to the new release, I opted for reinstalling from scratch while keeping my old home partition intact. I mainly did this as a way to clean up house, so to speak. However, this time around, I chose to perform a regular upgrade. Although by now there are many tutorials available on the Web explaining how to do this, I chose to follow the procedure documented in the official Debian 13 (trixie) release notes, and it worked fine. I'm now in the phase where I must run all my daily applications to make sure everything is working as expected. So far, so good. However, I noticed that the Telegram desktop application had been removed during the upgrade process. It didn't take long to find out why. As it turned out, the package itself had been removed from the Debian repos. I'm not sure why. I didn't bother to check. Going to the Telegram website and downloading their software for Linux was easy enough anyways. Next, I'll have to test and see if I can connect to the network at work, and perhaps migrate from the Nouveau drivers to the Nvidia drivers. I'm still not sure about that last one. In any case, that will have to wait until tomorrow. It's sort of getting late. {link to this entry}