Posts by Filippo

Telling internet platforms where to stick public service media will serve nobody. Turn it on its head

Your local newspaper decides what to put on the front page based on what they know about their target audience, and the reaction they want to get. That's literally what an editorial decision is.

I really don't see why, if you replace the editor with an algorithm, then suddenly it's no longer an editorial decision and most regulations stop applying.

It doesn't work like that in any other industry. If you automate a factory, you don't magically become no longer liable for product defects. If you automate management of a fund, you can't suddenly ignore finance sector regulations.

Why the fuck aren't social media platforms considered publishers, as soon as they base feeds on user profiles?

Boffins peg narcissistic leadership as the real driver behind 'return to office' demands

And unnecessary RTO mandates aren't even the only, or largest, damage from narcissistic bosses. At some point, we're all going to have to stare the productivity paradox in the face, and start asking hard questions about what the point of monitoring everything all the time really is.

Europe's chip ambitions won't break dependence on US cloud and software, says Forrester

Because if you can't do something in 5 years tops, preferably less than 2, then it's not worth even considering. I think about medieval cathedrals and weep.

Researcher poisons open-weight AI model for under $100

A typical use case involves models that have been "fine-tuned" or "distilled" or whatever the buzzword is to express the notion that you're not using the exact same weights as OpenAI (or whatever) published. A third-party has tweaked the model to make it work better for your scenario. For example, to get a much smaller model that can do a specific job just as well or better than the general-purpose model at a fraction of the inferencing cost.

If the third-party is malevolent and/or compromised, you have absolutely no way to know. Even if they share with you the training data and the entire procedure they followed to get the final weights you're using, you can't just reproduce it to make sure they're giving you what they say they're giving you, because running the training is extremely expensive.

Photovoltaics are still running after a year under Swiss trains

Don't get me wrong. The more clean power generation we install, the better. Keep it up.

But I don't get why some projects are so focused on finding specifically the toughest places to install photovoltaic panels. Remember the company that wanted to build roads out of them? That was nuts from day 1.

I mean, have we run out of good old boring roofs and parking covers?

EU puts 'addictive' design of Facebook, Instagram under the DSA microscope

Yeah. I'd like to have a separate term for "targeted by context alone" versus "targeted by user profile". Targeting by context is fine, it's been done since forever; I bet Romans hawked strigils in bathhouses.

Targeting by user profile, on the other hand, is like having someone stalk you. Just because they won't outright knife your back, it doesn't mean it's fine.

Re: Close...

Can't. It's not a sufficiently tight definition to base legislation on. You don't want to throw out babies with that dirty bathwater. There's message boards, messaging services, news aggregators, all kind of useful things that can be toxic but aren't necessarily and by design toxic like Facebook ended up being. "Ban social media" would probably kill the service I'm typing on right now.

It would save them so much money, because right now advertising is a near-monopoly, with all the distortions that entails. But anyone can serve untargeted ads. Kill targeted ads, and the monopoly is gone.

I don't think they "failed to assess [the risk of addictive design]" as much as they assessed it, decided it was a good thing, and made sure to have as much of as possible.

Also, I don't think the current design of Facebook (and most other social media) is salvageable. Does anyone really believe that turning autoplay off by default and popping up a "you've been scrolling for X hours" message every now and then is going to fix the problem? Or even make a noticeable impact on it?

Fix the root issue. Ban targeted advertising. We had untargeted advertising for centuries, we had it on the Internet even; it works just fine. Banning targeted advertising would, in one stroke, sort out so many problems it's not even funny, and it has literally no drawbacks for anyone except a handful of big tech companies.

EU 'Chat Control' snoopfest returns after vote to kill it falls short

Re: Chat Control.....

In a world where snooping is normalized, anyone who behaves in the way you describe will be put on a watchlist.

You cannot fix a political problems via technical means.

Europe's new import rules are coming for your bargains

Re: Seems like nothing more than a money grab

>Surely that can't be morally right?

I know I'm going to get downvotes, but... the answer would be, it can be morally right if the main reason the item is so cheap is that it was produced immorally (e.g. under horribly unregulated workers or environmental conditions). This is sadly the case for a lot of these items.

It's still annoying as hell, of course.

Apprentice developer defied orders – then got a job supporting her weird code

Programming is like sex

One mistake, and you're stuck giving support for 20 years.

Failed blockchain project ends with big fine for fibs about it being on track

The rule of thumb for blockchain projects should be:

1) Ask why you're not using a regular DBMS.

2) Absent a valid answer to (1), do not use blockchain, use a regular DBMS instead.

User swore hacker called General Failure had invaded his PC

Re: Any key?

I got into a couple screwups due to some alert popping up on foreground while the screen was off, and me waking up the screen with "space" or "enter". I've since gotten into the habit of waking up screensavers with CTRL or SHIFT. I've no idea why system developers don't have the screensaver just eat the first keypress.

AI bills are baffling the C-suite after shift to usage-based pricing

I have a solution that will allow your company to get any kind of work done at a fixed, predictable cost. My solution not only covers those fuzzily-defined job items that LLMs were supposed to automate, but it excels at them.

It's called "hire someone".

I'll collect my exorbitant consulting fee now, thanks.

Nearly a third of corporate leaders report difficulty understanding

The article could've stopped there and would still have been accurate.

They've read the scroll thing! AI helps decipher ancient document charred by Vesuvius

I wish everyone would stop conflating "AI", "LLM" and "machine learning". They are all really quite different things.

Small island nation tries bold tech education strategy

Re: Have you met teeanagers?

Liam has helpfully provided a link to exactly what he proposes, but it appears that most commenters aren't bothering to click and/or read it.

The one-word summary is: moderation.

The multiple-words summary is: age-based restrictions are not a solution, they will not work. A candidate solution that could work would be removing anonymity and forcing platforms to moderate contents. With actual enforcement.

This would reduce their profits, so it would definitely be a fight, but it could be done. The worst they could do in retaliation would be geoblocking the UK, which they won't, because they know that it would result in a strong local alternative popping up within six months to a year, and then expanding from there.

Waymo hits the brakes after robotaxis keep missing the signs for freeway construction zones

Re: Guess you haven't read articles / seen pictures

>Ok what about an Army checkpoint in N. Ireland? I'm supposed to just drive straight through because they aren't an authorised traffic officer ?

Well, there's a bit of a philosophical point there. Where does authority ultimately come from? The endgame of confronting authority, after (hopefully) a lot of steps, is getting cuffed and thrown in jail - i.e. the state monopoly on violence. In that sense, the Army checkpoint is most definitely authorised.

A modest proposal: Reformat everything to make documents more palatable to AI

I don't get it. Wasn't literally the whole point of LLMs the ability to digest unstructured natural language documents?

Feds freaked over Fable 5 after simple 'fix this code' prompt, not jailbreak, says researcher

The supposed model capabilities are just an excuse. Everybody who plays with these things knows that the top models are roughly equivalent. The point is that Anthropic is currently on Trump's shitlist, and therefore the government will gladly abuse its power to screw with them.

This is your BIOS speaking. Please fix me. Your PC is broken

Re: False economy?

The false economy is assuming that building PCs out of parts saves you money. That's only true for fairly small fleets. Beyond a certain fleet size, not very big at all, the savings from standardization eclipse the savings from reutilization, due to economies of scale in both maintenance and procurement. The other AC post goes into more detail.

Devs know AI code is riddled with holes, but ship it anyway

In fairness, the concept of "I know this code is crap, but I'll ship it anyway" long predates AI.

'Please do not vibe f--- up this software': Broken backups spark AI coding row in rsync project

All of those things are fully deterministic, explainable, and work to a specification.

That an app 'Fits on a Floppy' is still a useful measure in 2026

The lifetime of a bridge is measured in decades to centuries, with maintenance every year or so. The lifetime of a bridge building technique is measured in centuries to millennia. A bridge that requires less maintenance is considered a superior product. A building technique that works through the ages is considered fundamental. Physics don't change, ever.

The lifetime of a typical app is measured in months to years, with a patch every month or so. The lifetime of a language or major framework is measured in years to decades. An app that doesn't get frequent patches is considered dead and to be shunned, and the same goes for a framework or language that doesn't get new features. Fundamental architectures aren't stable for more than a few decades.

The people working on bridges, however, are still people, the same as the ones working on software. They can learn techniques, design products, and improve methods at roughly the same pace. You see the intrinsic problem?

Real engineering is the product of generations upon generations of accrued knowledge. You just cannot do that for software, not right now. There are a few timeless lessons, and coders would do well to learn them, but the vast majority of it is far too fluid. I can come up with a beautifully simple and functional piece of software, sometimes, but I shouldn't have to; it should be a solved problem, in hundred years old textbooks. There's very little of that in software, and not nearly enough to build a word processor out of it.

I don't have a fix for this. I hope that a few more centuries will be enough to make software engineering into a real kind of engineering.

FCC warns US broadcasters their licenses are a privilege, not a right

Okay, as long as we're clear that...

"Public interest" is neither "Government interest" nor "Trump interest".

Bezos rocket fell short after cryogenic leak cut engine thrust

Re: Too many failures

It's not just the US. Generally speaking, I see a spreading mindset which believes that the world somehow owes us to be simple. Pointing out that something is complex and cannot be reduced to a soundbite appears to be outright offensive to this kind of mind. How dare you require effort of me.

Given that reality owes us nothing, doesn't care about offending or being offended, and sure ain't getting any simpler, this will not end well.

Re: Too many failures

I am very concerned about the general public's inability to understand what "success" or "failure" means in any context that features any degree of complexity.

Pope Leo warns AI boom can give Big Tech and the people who run it too much power

>A more moral AI is not enough if that morality is determined by a few

That's rich coming from the Catholic Church, that well-known bastion of letting the many determine morality.

No captain, my captain: Navantia floats crewless warship

Re: Autonomous and armed

I believe he meant, besides the target.

Gemini accused of 30,000-line code purge and fake recovery report

>the model broke core functionality, made sweeping unrelated changes, and left the system in bad enough shape that the changes ultimately had to be rolled back.

That sounds like quite a lot of wasted time. I wonder if it will factor into those "AI improves dev productivity by XXX%" reports, or if the report's design will be carefully designed so as to ignore it.

America's top cyber-defense agency left a GitHub repo open with passwords, keys, tokens – and incredibly obvious filenames

Re: Ho Hum !!!

Well, the thing is, if you run a honeypot, you don't want the bad guys to figure out that you're running a honeypot, or even that you were running a honeypot until yesterday (otherwise, they'll be more suspicious of future honeypots).

So, in this case, an amazingly competent CISA would look exactly the same as an amazingly incompetent CISA. Each of us can pick their favorite version, happily call the other camp idiots, based on the same amount of evidence.

Frustrated franchisee sues Pizza Hut over crappy kitchen AI

Re: it's a good news story for the drivers

What? No! The point is for shareholders and the C-suite to game the system and be lauded for it. If an employee or - God forbid - a gig worker attempts to do the exact same thing, they are to be accused of being the reason the system fails, publicly shamed, and, if at all possible, prosecuted.

'Big AI' is subverting regulations just like tobacco and oil firms

Not surprising. Finding a reliable way to deal with regulatory capture would be a great achievement of political science, if it ever happens.

Anthropic urges Uncle Sam to kneecap China's AI ambitions before 2028

Re: Digital sovereignty in Europe

OP is talking about software. You're talking about hardware.

Yes, there's significant overlap. Yes, there are some aspects of sovereignty that cannot be had without control of both.

No, they are not the same thing, and answering a post about software with objections about hardware will only generate confusion and communication failures.

Dissatisfied: Three-fourths of AI customer service rollouts are a letdown

Re: surprise

That's what the remaining one quarter are doing.

Sovereign cloud is only possible if you’re Chinese or American: Gartner

The entire argument is based on a wrong premise - that the cloud market is perfectly globalized.

Case in point: the argument starts with impossible [...] outside of China or the USA but then continues with [...] it predicts the cloud market has settled around AWS, Google, and Microsoft

You might have noticed that none of those are Chinese. That so-called "rule" does not apply to a segmented market.

The market is already not-globalized because China is serious about sovereignty, and the EU absolutely could decide to get serious about sovereignty, and just segment its market out by law. It might not happen, and it will be slow if it does happen, but stranger things have happened.

At that point, European sovereign cloud services would appear, painful as it might be.

Musk has never built a wafer fab, but he wants to burn $119B on one anyway

Re: What?

>Shriveled husk? Compared to what? Bluesky and Mastadon??

Compared to Twitter. I mean, it's right there in the same sentence.

Palantir CEO: 10 percent of the world 'professionally hates us'

Either that, or it means I can get paid for hating Palantir, in which case I'd like to know how to collect my check.

Re: Err... about that schoolhouse...

Also, I'm just saying, but if you're naming your company after a corrupted artefact that will drive everyone who uses it insane... well... one should ask oneself a couple questions before doing business with you, right? I mean, that's only marginally more subtle than calling yourself "Evil Corp".

So, I understand that "professionally" is the new "literally"?

Pass the key, passwords have passed their sell-by date

Re: Great!

That's not a passkey. That's a password manager. It's literally what I described.

Re: Great!

I understand that passkeys add a strict dependency on a specific device, an Internet connection being available, a specific cloud service being available, or a combination of the three, depending on implementation. Any of these is a point-of-failure, which can and does fail in real life, and whose mode of failure is that you're locked out of the service you want, with no easy workaround.

By comparison, a proper password manager with encrypted local storage and sync across several devices lets you use high entropy passwords that are immune to guessing and bruteforcing, and only locks you out if all of your devices are unusable AND you're offline on top of that.

If I'm wrong, I'd like an explanation or a pointer to where I can get one.

Intel eases reliance on TSMC with 'Merica-made Core Series 3 processors

Re: Core 3?

I have a fanless PC that sports a Celeron N5100, which has a cpubenchmark.net score of 3311. It runs Windows 11, and I mostly use it to run small light-load server programs for various purposes, or to test stuff. When I need to use the GUI to install things or diagnose something, the performance is adequate. That's Windows 11, so not exactly a champion of efficiency.

From looking at cpubenchmark.net, the worse Core 3 I can find has a score of 7353.

By and large, nowadays, if what you want to do is check email, browse average websites, and do some office tasks, you would be hard pressed to find a CPU that does not do it well. If a Core 3 is struggling at office tasks, either they are not really office tasks, or there's some problem.

Opsec oopsie: Dutch navy frigate location outed by mailing it a Bluetooth tracker

Yup. There's some device on the ship that's Internet-capable and unsecured. That's the root problem.

Google Chrome lacks protection against one of the most basic and common ways to track users online

I'm convinced that we've been approaching this problem from a fundamentally wrong angle. The reason I don't worry very much about being tracked everywhere when visiting a shopping mall isn't that I only go there with a full face mask; it's that the shop owners are outright forbidden from uploading their security camera footage to an ad broker.

Similarly, browser features that enhance privacy are a great thing - but I don't want to "win" the arms race; I want to not be in an arms race to begin with.

Just pass legislation to ban tracking. Off the top of my head, I'd just state that everything that an endpoint sends you, including the originating IP address, is personally identifiable information as per GDPR. It can't be shared with third parties, and you can't use this as an excuse to deny service. Build tightly defined exceptions for CDNs and VPNs. Maybe for SSO services, I guess, although personally I wouldn't be sad to see them go. Those things can work, but they have to keep data siloed. Yes, many other things would break (among which the business models of most social media), but guess what, in most cases, they were only designed that way in order to track you, and can be redesigned so they don't need to.

There are probably problems with that particular definition, but the point is - we all know why the big techs do this. Let's stop debating the how. Kill the fundamental reason.

QUIC will soon be as important as TCP – but it's vastly different

Re: elephant in the room (or web page)

Well, yes, but that's not really a problem for computer scientists. We don't complain about gridlock to car engine designers.

Don't let the bot play doctor! AI gets early diagnoses wrong 80% of the time

In fairness, machine learning might have a place in diagnostics. Wasn't IBM trying this with Watson, well before ChatGPT was released? The idea was that, with enough data, you could calculate the conditional probabilities of each possible disease from a given set of patient characteristics, symptoms and exam results, then calculate which exam would have the largest discriminating effect, prescribe that, and then iterate until you had a candidate diagnosis with high probability. This wouldn't replace a human doctor, but it would be useful to warn a doctor to consider an uncommon disease in cases where symptoms are uncertain.

It would also work off well-understood statistical methods, rather than black box LLMs. If I recall correctly, they couldn't get it working because most medical knowledge is encoded in natural language only, making it intractable by traditional methods. They might be able to use some specialized LLM to ingest medicine books, papers and whatnot, and turn that into structured data to feed into a traditional algorithm. The LLM will get something wrong, but it's all statistics anyway, so there would be some built-in tolerance for errors.

But using a chatbot directly to make diagnoses? No, just no. Remember, it was trained on the Internet, including every single nutjob with a keyboard. No.

Unless they're absolutely sure, a human medic will almost always tell you that they don't know yet and need to run some exams. I'm not sure whether you would count that as success or failures, but it's certainly a healthy approach.

AIs, on the other hand, rarely tell you they don't know, and tend to start spouting unsupported hypotheses, often while projecting strong confidence.

The only technology that died more times than VR is AI, and that seems to have worked out

There's a bit of wobbly logic at the end of the article.

The author argues that what we lack is a "killer app" for AR, and brings as examples Roblox, Minecraft and Fortnite.

However, all of those are fully VR worlds, and none of them are AR. Although they use similar tech, VR and AR are extremely different domains, and you really cannot conflate them if you are discussing applications.

VR has failed and continues to fail largely because isolating yourself from surrounding reality comes with a large number of extremely difficult challenges (both social and technical) that have nothing to do with how good the visor is. Let's pretend you could stuff it in lightweight glasses, make them 16K full FoV resolution, and make the battery last 24 hours - guess what? Most people still wouldn't use them, because almost none of us can have a "VR room" in their house. And if the VR app doesn't require me to move around, I might as well just sit in front of a screen: it's not actually much worse, and I can use a coffee mug at the same time.

As for AR, that eliminates the problem with your surroundings, but the problem of "do I really want to move around, and if I don't, why don't I just use a screen" remains. That makes the space of interesting applications very limited: I'll reiterate that not a single one of the author's examples is AR. There is exactly one AR application that I would call very successful, and that's Pokemon Go. As far as I can see, it was a one-off, never repeated, and it worked just fine with a regular smartphone. It's not exactly looking good for visors.