Krssst

Can run on modern 64-bit windows too with otvdm: https://github.com/otya128/winevdm

Another way to think about this: if the compiler has a miscompilation, the user is usually not blamed for the mistake (unless gross lack of QA). If the LLM makes a mistake, the user gets full blame. Thus, slowly combing over the output is required.

If we get precise detectors and LLM posts don't get shown by social networks recommendation algorithms as a result, the chances of people starting to talk like LLMs get lower.

We can measure false positive rate. The detector in the arricle is 85% accurate (not sure about false positives, but let's assume) which is too low to make conclusions, but enough when browsing the web and skipping reading likely-slop withiut accusing anyone.

If the false positive rate becomes <1% then it's better. The alternative is the world drowning under slop so I'd rather have imperfect detectors and have users aware they may fail in rare cases to avoid witch hunts. The general issue is that people only realize they're reading slop halfway through which is frustrating. If you know it from the start thanks to a detector and move on without commenting, no time waste, no frustration, less negativity towards LLM users.

The article mentions that AI texts are often caught by multiple models, so hopefully text from newer LLMs could still be caught without updating the model?

The classifier does not seem so big, I wonder if something like it for English could be used in a browser extension to run against every single paragraph being displayed ?

If the internet is going to drown in LLM text it would be nice to have tools to detect that automatically just like we have adblockers today to avoid wasting time on ads.

(the article was a good read, thanks!)

According to 3 out of 11 tested detectors, the other 8 giving a human rating. The only reasonable conclusion is that those 3 tools are broken, cannot conclude for others. I'll keep trusting Pangram.

It's good to have this kind of data in any case. I assumed it was easy to measure accuracy with test sets for this kind of ML problem, so the detectors cannot be too wrong if evaluated properly, and that seems to be the case for the 8 non broken ones.

Yes, I dislike this kind of take so much. It keeps being repeated as a truthism and a way of putting down people that don't do what the speaker wants. It's fine to disagree, but there's no need to get such a threatening tone.

A lot of tech jobs seem to be only about sheer output volume, with quality (maintenability, availability, security, generally understanding what the thing is doing) not mattering much. In that case sure, LLM all the way and whatever happens happens. But not all jobs are like that.

If people develop long COVID after catching COVID (even before the vaccine existed), but not after taking the vaccine, then the cause is the virus not the vaccine.

Those weird comments coming back again and again is just insane. I wonder why does HN attract that kind of people.

Oh, it did not feel AI but Pangram does say AI, high confidence. Good catch.

(I'd trust an ML algorithm that only has to classify in two boxes, thus easy to evaluate, over my own gut feeling).

As much as I disagree with the general consensus that the article follows of "delegating non-automatable work that requires thinking and understanding to the machine because it is boring, even though the machine is unreliable", according to Pangram this does seem human-written (confidence low).

AI detectors are criticized but classifying stuff into two boxes is probably one of the stuff that is the easiest to measure the accuracy of (as long as one does not put the test set in the training set...).

(well one could see the irony of using ML to detect ML text while complaining about people not caring about understanding anymore, but that's one case where the machine is more reliable than the human)

Vaccines were very effective against the first variant, and got less effective with later ones. People forget about the timeline. Article mentions the delta variant at which time vaccines were still very effective IIRC. There were some breakthrough cases as the article mentions but that's to be expected with anything short of 100% efficacy.

Maybe I’m a bit weird, but I like seeing all these warnings: “You have unsaved work! Are you sure?”

One of the reasons I don't like to reboot: Windows taking a few seconds to show the warning, so I have to either babysit the computer until it shuts down or come back the next day with an unrebooted computer.

Now I just use shutdown /f to force shutdown/reboot and forget about it.

If Chat Control gets through, it means the Parliament approved it, which means the EU people voted for politicians that supported the idea. If the EU got dismantled, the same politicians would be elected (they won once why not twice) and do it again at the local level. (though, maybe not in every country)

Doesn't wine have various rules to remain a white-room implementation?

Not sure using LLMs which have possibly been trained on leaked Windows sources would be compatible with that. But that's just speculation, I wonder if LLMs possibly using leaked sources for training has been looked into. (probably legally difficult as the investigator would have to access the leaked sources too...)

Using Windows Server 2025 (it has an evaluation version), I encountered a few problems (Xbox Controller needing to dig out de Windows 7-era drivers, Meta Quest audio not working, using pnputil to import missing drivers from a normal Win11 install) but otherwise it's been quite smooth sailing.

Not having Store login sounds bad, but it also means the system cannot trick you into linking your account to a Microsoft account, which is a plus (though accidental login is reversible IIRC). (I am not sure if Minecraft, which is the only game I know to require such login, actually worked or not).

Using not-for-purpose OS for gaming does lead to some hiccups, but to me those hiccups are preferable to the constant fight against your OS trying to shove things down your throat or disregarding your choices (of not wanting copilot, of wanting a local account, of not wanting ad-like stuff in the OS).

(Fedora would be easier to setup at that point, but anticheats...)

I also tried the IoT LTSC evaluation which generally worked better (basically, it has all the drivers the Server version is missing, plus QoL features like Win+V are enabled by default) but buying legitimate keys was not possible as a regular consumer.

Being sick ~20 days per year was the norm for me before remote work (and using N95 masks after it ended). Some people are lucky, others are not. (well, "luck" actually encompasses many factors such as population density, presence/lack of children, mandatory office presence, proper infection countermeasures in public places (air filtering, aeration), and I assume sufficient sleep or not (the only time I did not get sick as often without masking was when I managed to sleep 8 hours a day for a few months))

I'd be curious to see how much the decision to not recommend it in older people is "cost / benefit" ratio (the vaccine is expensive and if you say it may be slightly useful you might be putting pressure on yourself (as a government agency) to reimburse it even if the money would be better spent elsewhere) and how much is "actual risk (including getting hit by a truck on your way to the doctor) / benefit" ratio. I don't know of significant side-effects of the HPV vaccine, though my understanding was that there was some misinformation going around when the first vaccine became available which may have made governments more cautious regarding rollout. I don't know much at all about this topic however so this comment is likely full of mistakes.

Fun fact: on modern Windows, if you uninstall modern notepad using the Settings app to get the old one, you won't be able to associate .txt files with notepad.exe without a registry trick: https://superuser.com/questions/1750222/how-to-open-file-typ...

No idea about the overall background of the project or devs, but: https://github.com/X11Libre/xserver/blob/master/CoC.md

In my understanding a good chunk of the CoC dislike connects to MAGA ideology (because CoC aim at inclusivity, and we know what the MAGA position on that is).

(the standard approach to thinking a CoC is useless is probably just to not have one and not comment on it. Going out of one's way to make a statement shows something that goes further than "maybe it's useless")

The alternative before is that you knew where you were reading the information from. So you knew what level of trust to give each information.

AI summaries removes that ability from you, and even when it gives the source it may paraphrase it incorrectly just because LLMs are fundamentally unreliable. The level of trust to give LLM summaries is 0.

There's much more air outside than inside, so 15C colder inside does not mean that the entire city gets 15C hotter outside. And in a heat event, most people are inside, not outside. 1C hotter outside to make it livable for 99% of humans sounds fine. And this is only about cities, anything living outside cities will be fully unaffected.

For the people that have to work outside: air conditioning in the vehicle, frequent breaks in air conditioned areas, and I wonder if we could get proper air conditioned clothing at some point (currently vests with fans embedded are quite frequent in Japan, but that's the best there is as of today).

But I agree with the last paragraph. Air conditioning is the only countermeasure we have but in the end the fact remains that many cities will eventually become incompatible with human life in summer.

The cursor animation is actually a great one because it does not add any latency. By comparison, when animations are not disabled on my Pixel 6 it takes almost one second to switch application instead of maybe 100ms (double tapping the app swap button to get to the previous app running).

What are those dark patterns? It's an off button, it works, and it does not get back on. It's the polar opposite of the "maybe later, I'll ask again every week and reset the setting in your back" unfortunate norm that plagues a lot of major proprietary software/service.

Games are for fun. Wasting time in a game is fine, that's what it is for. (edit: not saying that pejoratively)

Other applications are to do things. They should do the thing and get out of the way as fast as possible. Animation-induced delays are fundamentally contradictory with that; they waste the user's time instead of doing the thing.

Firefox will also disable V2 sooner or later.

Source?

Firefox won't, because mozilla banned that extension from store.

It's unbanned; the author chose to not put it back. https://www.ghacks.net/2024/10/01/mozillas-massive-lapse-in-...

France is quite likely to put the far-right in power next year, don't get your hopes up. I wish it wouldn't happen but a lot of people seem like they will not vote in a runoff between a far-right candidate and a left-wing (too left for the regular right) or right-wing (too right-wing for the left) candidate. It's absurd but not much one can do about it...

More than non deterministic : LLMs don't have a specification to obey to in the first place, while compilers (rather, programming languages) do.

Fully agree with the article (as my comment history would attest). I wonder why people worry so much more about "determinism" over conformance to a spec. (a compiler can be nondeterministic and correct which is not particularly good but definitely not worse than being incorrect)

Maybe most devs don't care so much about the language specification and just expect the code to do vaguely what it looks it should do intuitively? This is not very clear to me, at least for libraries I guess a lot of people don't read API docs and just call the API hoping it does what the name says (preconditions be damned) in which case nondeterminisc observable behavior would be more problematic than nonconformance to a spec (API doc) they don't read.

There will still be jobs. Manual jobs, the kind that break our backs and have us breath various stuff we shouldn't (dust, fumes). Robots are difficult and maybe not so economically viable when everyone is desperate for any job at any cost.