GuB-42
And if you wrote a blog, these would be the events you would write about. You obviously won't write a blog post about events you don't remember.
It is selection bias, it is not indicative of the ability of Raymond Chen to remember all little details.
It is like counting tolls as a component of "employee compensation" for a driver. An employer may decide to have a driver use toll roads, which may result in faster deliveries and more efficient use of his time. From the employer point of view, he is paying more to get more work done, from the driver point of view, he is paid the same for the same amount for time driving.
From my experience travelling in Japan, online booking is a mess. But every time, the solution was simply to go to the station, walk to the counter and ask. The guy will get you exactly what you need, may give you alternatives if what you want is unavailable, tell you about unusual connections, take into account whether you prefer to be seated together or if you prefer to leave ASAP. This is real service from competent tellers. And I am saying this as a tourist who practically can't speak Japanese. And for paying, cash always works. Alternatively, if you don't need to talk to someone, there are also some ticket machines that are rather well designed.
It turned out to be a better experience than in most other places that may have better websites and more digital payment options. Yes, you have to physically go to the counter, but the is a good chance you are going to be in train stations a lot, it is not really a problem.
It is millions of time better than some booking websites that are designed to extract as much money as they can from you instead of helping you (ex: low cost airlines). No, I don't want insurance, no I don't want extra luggage, no I won't pay extra to choose my seat, no I don't want priority boarding, no I don't want to rent a car, no I don't want a hotel, no I don't want your app, no I don't want to subscribe to your mailing list, yes, I know what I am not getting, don't try to upsell me a second time, I just want a ticket to my destination!
We may imagine the CEO of a large company in his throne room, only reachable through his servant and only for matters of utmost importance, billion dollar contracts and all that. And sure, there is that, but not just that.
You may occasionally find out that the guy who a replaced a defective Ethernet cable on the intern's computer was that CEO. Or the lead developer of a AAA game patiently explaining some basic functionality to a random gamer. A CTO may pick up a random support call. So Raymond Chen debugging an "obvious" problem in a .bat script, not at all surprising to me.
It has already happened to me that on a support call where I expect some underpaid grunt, find that the guy is really good, patient and friendly, only to end up discovering that I just talked to the one who built the whole thing. Makes my day when it happens.
It think that if I was some kind of VIP top level whatever, I would cherish these moments. Simple things, genuine human contact, away from the big responsibilities tied to my position. And even from a professional perspective, getting a feel for what happens down there is important.
For example, by distributing cracked software, you compete against the official channels, so, not fair use. Quoting a book in an article won't make the people who would buy the book go with the article instead, so, fair use. AI companies argue that training is fair use because their model is a completely different product than what they are scraping. A LLM that has read the Harry Potter novels is not a substitute for the Harry Potter novels.
But all these rewrites make the fair use argument flimsy. These models are used to make knockoffs of the products that are part of their training data set, at least in part. Not fair. It is in their best interest to limit the amount of knockoffs their models are used to produce. Imagine if they had to compensate the people they took the work from...
In addition, if fuels their argument of "we are making sure our models are only used to make safe and innovative content, unlike these bad open models that are used to make dangerous knockoffs".
Short words mattered in the early days, with punch cards, very limited memory and computing power, shorter words saved resources. Also, when running a program means handing a stack of punch cards to an operator and waiting for the result to come out the next day, saving a few seconds of typing probably wasn't a priority.
As computer got fast enough not to care about single bytes, there was a push against using shorter names. "be descriptive" they said. Also we had more advanced editors with copy-paste, search and replace, etc... If there was an objection for overly long variables, it was more about screen real estate than typing speed.
I'd say the case where it mattered would be interactive shells, UNIX shells in particular, because for DOS, commands were limited to 8 characters by the filesystem, another technical limitation. But then came completion, making typing speed not really an issue, but you still want to be able to fit your command in one line, because otherwise it is annoying.
So: technical limitation or readability rather than typing speed.
They have top of the line research, I mean, they invented the thing, and they are still doing stuff that goes beyond making a bigger LLM, as it seems to be the only thing the others are doing, research-wise.
They also have hardware capabilities no one else has, they invented the TPUs, and have lots of them.
They also have the brand recognition, ecosystem, user base and lots of money.
So why are they lagging behind on the actual products, and by lagging behind? I mean they are always in the position where they are just behind the frontier, but still in the game.
It seems almost deliberate. Maybe they are playing the long game, waiting for the bubble to burst. Maybe they don't want to compete against the more profitable parts of the company. Or maybe their researchers are not interested in making an actual product unless being forced to.
And it is weak because people are different. Some people tolerate clock change better than others, some people function better in summer time, others function better in winter time. Any decision you make (keep clock change, summer time only, winter time only) will benefit some people and will hurt others. And there is not even strong evidence that one solution is better on average.
For the disruption caused by clock change itself, it is indeed noticeable, like an increase in traffic accidents, but it is just one day, but the effect of being in summer time vs winter time lasts 6 months. So if the clock change makes even a small improvement for 6 months, it may be worth the disruption for 2 days a year. I am not saying there is, it is just that there is no real consensus.
So, by default, when a change may have negative consequence for some people and may not lead to an overall improvement, you don't change.
Some people reject AI on a matter of principle. They may have a problem with the energy cost, the idea that AI companies are stealing the world knowledge for their own profit, that any use of LLMs will make you stupid, or that it does more harm than good overall. The code of conduct links to a short story by Isaac Asimov, you typically don't do that when your motives are purely technical.
To me "no LLMs for finding bugs" means what it means, which implicitly means "we prefer to have bugs than for you to use a LLM". It is weird, may result in worse code, but it can be justified for ideological reasons, and I mean, why not, it is your project, your rules.
In the unlikely even I contribute one day, reading this rule, I will turn off my LLMs completely and my submissions will probably have bugs that could have been avoided using a LLM based debugging pass, but that's the rule, and I won't discuss it because it is also the rule.
If the intent is to avoid flooding maintainers with bug reports, say "no LLMs for bug reporting".
It is common for well meaning people to give medical advice even though they are not doctors. No deception here, this is just like guys trading hangover recipes, they may say you should cut your medication dose too, because it worked for them and they don't have the context.
Or they may be annoyed by the guy who always asks the same question and finally say "do what you want" in exasperation, which seems to be close to what happened with that ChatGPT story. Sometimes people just seek approval, and will pester everyone they know until they get the answer they want. Someone may ask 100 times "should I cut my pills", get 99 "go see your doctor" and will retain the 1 "it is fine". Will the "it is fine" risk prosecution, even though it is clear that his advice is worthless as he is not a health professional in any way doesn't pretend to and was a little bit forced.
Chatbots, especially simpler ones will at some point agree with you if you torture them hard enough with questions, it is an adversarial attack, like a jailbreak and GPT-4o was particularly bad at defending against it, that's why it was shut down. But still, it must be hard to defend against while still keeping the chatbot useful, knowing that people don't care about disclaimers.
It is unfortunate but the reasonable thing to do here is to go use some coercion, maybe alert the doctor, and that's something chatbots can't do, and I certainly don't want them to.
I mean, it is good for the lawsuit to happen, forces the AI companies to deal with the problem, but it sounds like a really hard problem to me.
Most of the links in the first page give the actual recipe, including the Wikipedia page. The AI summary just tells me how bad it is.
On the other hand, giving the election results in the US is not illegal (obviously) and Google fixed it (at least for me), giving the the expected factual answer.
Edit: Not actually sure about there being a law relating to Tiananmen Square, subversive speech is illegal, but would DeepSeek break the law by talking about it? Not sure, but it probably could get them in legal trouble.
But here, it probably saw that 2020 elections made a lot of people angry, with riots and all that, and thought, "nope, I am getting out of that one", and it took over what would have been a straightforward answer.
It reminds me the time an AI (maybe Gemini, not sure) depicted a "diverse" US senate in the early 1800s, a time where it was all white dudes. People got up in arms over this.
My take: it is just a bug, move on, at least give them the time to fix it before getting outraged.
Just look at the products these companies have created, the common theme being plastics, they essentially defined the modern age. We sometimes call eras by the material that defines it: stone age, bronze age, etc... We are now in the plastic age: look around you, what doesn't make use some kind of synthetic polymer. Even if you actively seek to avoid plastic, you are going to find gaskets, electrical insulators, coatings, etc... and electronics of course.
If it wasn't for these companies, we would be living like we did 100 years ago, which is easy to romanticize, but look at life expectancy during that time, so, no thank you.
Did they do bad things? Of course they did, very bad things even, leaded gasoline comes to mind. But overall, I would rather be in a world where these companies exist than one where they don't. But for the tobacco industry, a world without them is, I think, unquestionably better.
Meta and Google? I'd say it is like the chemical industry, but on a lesser degree, they do good, they do bad, but overall a net positive. It means I agree with your second sentence, though in a different way.
To me, the name "Resource Acquisition Is Initialization" implies that doing one means the other one is done implicitly. Now, in the end of the day, the result is machine language, with proper discipline you can follow the RAII principles in any languages, including assembly, but I have always understood RAII as being it done implicitly.
Also we can argue that Zig supports a weak form of RAII through the use of "defer", there are some nuances here.
Languages have opinions, that's why there are several of them and they are popular in some applications and not others. And Zig opinion is to be explicit so RAII is out, and the language does push you into avoiding allocation when unnecessary, as a side effect of not giving you the "easy to use interface" you want.
This makes it particularly suitable for fields like embedded software where you may not have malloc/free at all, but may be less so in a highly dynamic environment, where you may be better served with a GC anyways.
Nurses and doctors don't refuse giving painkillers because they enjoy seeing their patients suffer, painkillers come with side effects. So they are trying to find a balance between preventing suffering and long term health. We can look at the wrong idea that black people have a higher pain tolerance as causing black patients unnecessary suffering, but you can also see it as protecting the long term health of black patients more. That's the same thing, just seen from a different angle.
But now that we know of this bias, we can of course try to fix it, but how. Now that we also have data: do black people have less painkiller-related problems? If they do, maybe we should give the black treatment to others, if they don't maybe they could get some more.
I am talking about professors taking the work of their students as their own, unpaid of course, and the students can't say much because their future depends on it. So before calling simple copyright infringement "slave labor", they should look at their own community. At least the professors whose work is illegally copied are not forced to do it by the ones taking it.
And we here stories of "guy left big company to make his own startup" all the times. It is a common story for indie video game devs for instance.
And that's how that "layoff Discord" can produce something that can matter to customers. It will be a smaller company, and small companies have a serious advantage when it comes to support: you can get an actual human, maybe even the main dev, he can actually do things rather than transferring you from chatbot to bureaucracy. It doesn't scale, but it may be enough to get a competitor started.
As for connections, you may have inadvertently let go of the guy who has the connections. And you may have fired a perfectly functional team, ready for from clever entrepreneur with connections (who may be one of the fired) to pick up. It is so much of a recognized problem that some big companies hire people just so that their competitors can't have them.
Same idea for that transformer, it needs an infinite context, and it is an infinitely big machine, but with that level of abstraction, it processes a token in O(1).
In practice, there is no such thing as O(1) on arbitrarily sized data, there is no such thing as infinity either, it is just an abstraction.
The idea is that instead doing things correctly, you do "something", then, if it fails (it will), you change some random stuff until it passes. It typically results in code no one understands and that fails each time something unexpected happens.
AI makes the process much faster, but just as wrong.
Anyways, most of the times, SAT can be solved quickly, but not always, so getting the answer from Astra in polynomial time doesn't prove much unless you can also prove that it always does, no matter the input.
Modern day Furbies?
Hopefully, all these companies will go back to reason in a few years, when the bubble will burst. By bubble bursting, I don't mean that AI will disappear, but the nonsense and crazy money around it will.