Posts by doublelayer 10509 publicly visible posts • joined 22 Feb 2018
Sure, but that is an aspect, not the defining trait. Two people can have money and power and nonetheless use those things to do two completely unrelated, opposed, or perpendicular things. To describe us as different on that basis is logical. To describe them as the same based on that aspect is oversimplified.
It generally does, in that if you use a library that has existed and worked correctly in many countries for a lot of years, they probably considered time zones and leap days. It's usually not too hard to find something to help with time. Most programming language standard libraries and operating systems have that handled. Unless you have something they can't handle, the chances are that you will not benefit by either writing your own or trying to find someone else's library for the task.
If you're going to do so, perform the following basic tests:
1. Look through their documentation. If they mention oddities of time zones that they handle, they probably work. If they sound like students putting out something on GitHub, maybe not.
2. Check leap year information. Run this code or language equivalent foreach (int i in [2000, 2100, 2200, 2300, 2400]) print(is_leap_year(i))
If you get true, false, false, false, true, good sign. If you don't, don't use it.
3. Check what they did the last time some country decided to mess with daylight saving time for no reason. For example, you could see whether and how quickly they updated the time rules that Greenland changed in 2023. If they're using the typical sources of information, this could be automatic.
There is only one unless in their rule:
"It's a leap year when the year is divisible by 4, unless the year is divisible by 100 and not divisible by 1000."
2400 is divisible by 4, divisible by 100, not divisible by 1000. As I structure their statement, that does not allow 2400 to be a leap year. I don't see how you find a second unless, nor how you can make 2400 a leap year and 2300 not one using divisors of 4, 100, 1000, and 25000 singly. It's also just incorrect based on Gregorian rules. There is no rule for divisors of 1000 or 25000 and there is one for 400.
I'm not sure where you got this, but it is not correct. For example, the year 2400 is a leap year according to the Gregorian calendar, but using your calculations, it would not be one. Your rules also result in a different number of days per solar year of 365.24096. The Gregorian calendar's cycle is 400 years in length and repeats after that. There is no rule based on 1000 or 25000 years.
"Does it matter if December is in Winter or Summer[1]?"
It does to me. If, for example, we're tracking climate changes, we can ask a question like "how has the rainfall in December varied over the past hundred years" and get an answer. If December keeps moving between seasons, it's no longer going to work as accurately. In order to calculate it, you'd have to phrase the question like "how has the rainfall during the period from twenty days before the solstice to eleven days after the solstice" and do all the calculations. This isn't just relevant for months; we get the same effect if we calculate the rainfall between November 17th and January 3rd, assuming there's a reason to do it, because that or any other set of dates consistently refers to almost the same time in the solar year.
It would take significantly less time for December to become October, which would be relatively inconvenient even if it wasn't a complete reversal. However, even if you don't care about that, we get into philosophical areas about why we even have a calendar. If we don't care about the consistency between our date counting and seasons, why have months? In fact, why track years? Just count everything in days so you can give an age when necessary, and ignore everything else. I think we still find that having a way to describe, consistently, times in relation to solar movement, to be something we want to continue doing. Doing that inaccurately is not any better than not doing it at all.
I have not read the study, but I do wonder a few things about it. Specifically, how long they kept that up, because anyone who has stayed awake all night to do something knows that you can do something like that occasionally and be generally okay, but doing it too often has some really negative results. Also, the schedule is probably different depending on how much exertion or stress a person goes through, so people living in an experimental environment probably aren't doing the same amount of stuff as a typical person at a job, a student, or a parent.
If you're payed the same amount for every month, then you're getting more per hour or day in February, whether it has 28 or 29 days, because it doesn't have 30 or 31 days and every other month does. If it's something based on the number of days and they treat it as 28, then maybe you have a point, but I'd need to see that algorithm to understand if it applies or not.
I admit to being one of those who doesn't understand why we benefit from changing the clocks for summer time and thinking that we could manage equally well without doing it. Many countries have managed that. But I know why there are leap years and we cannot get rid of them. Most societies two millennia ago understood that and had figured out some method of handling it, so it should be pretty obvious how necessary it is.
And screwing up every human's sleep cycle. We're very hard coded for a 24-hour sleep cycle. Bad things happen when that changes too much. There are some people who have a circadian rhythm that's not aligned with that, but most of them are caused by total blindness where the light triggers used to form normal rhythms are unavailable. From online descriptions, the experience seems very unpleasant.
Yes, millennials can be blamed for that, as long as you're* willing to take the blame for everything that breaks in 2038. I'm willing to bet that that will be more things.
* I'm guessing that, because you said this, you're not a millennial and probably of the appropriate age to have been around when someone made the 32-bit signed time solution. You probably weren't the one to do it, but we're blaming generations for the actions for a member, so it's your fault.
Probably not. A lot of the companies making hardware with low margins outsource large chunks of the work to other companies that do it in bulk. This is how you end up with a thousand Android products that all use the same update server. This works because that update server only ever returns the "you are up to date" message. They don't have to modify the source as the thread above this one explains what was almost certainly the mode of infection. Whoever made the image, whether at this company or at someone they contracted with, included, either through malice or incompetence, some malware, and nobody checked it. If the person was being malicious, they were still incompetent because they couldn't infect it well enough to prevent it being detected and removed by Defender immediately on boot, so we're not dealing with a criminal genius here.
That is still probably a good reason not to buy from this company because they're making it clear that they don't care about doing anything correctly. I am willing to buy Windows machines from dodgy-looking companies for myself since I am comfortable wiping and reimaging them when received, but I wouldn't buy them for anyone else. This is also why I tend not to buy any Android devices from companies like this; I know what's likely to be in there and I do not have any reason to think I can get it all out. My rule is that, if I don't know where the update server is, it's an immediate rejection.
In that case, you are using the wrong tool for the job. A screwdriver twists things, a hammer bashes things, a search engine shows you websites, a generative AI returns what is most likely the result of a search of a large chunk of preexisting data with weights set by a prompt. None of those things is promising or even intending to give you an objective picture of reality. Don't expect one.
You can argue about what weights should be there and what should be in the training data, but no matter what you do with them, objective reality will not be guaranteed. Asking for a picture of a tree will give you a picture that is like the ones that were tagged with trees. Asking for a joke will give you something that looks like a joke, but it's probably either a preexisting joke or not actually very funny. Asking for the result of a calculation will give you some text that looks like it has run your calculation. If you try it, you will find that it often calculates something else entirely and gave you that instead, because it is a text predictor, not a calculator. Fortunately, if it's calculation you want, there are tools that are quite good at doing that. If you think generative AI is doing something it was not built to do, you will continue to be disappointed.
What gave you that idea? I don't just mean about diversity, but about anything? These models don't exist to give you reality, but to give you what you are likely to want. You ask for a picture, they assume you want one tailored for pleasantness or picturesqueness, not realism. Its training data has a lot more pictures tagged with "tree" of trees people thought looked nice on a background that was pleasant rather than a dead tree on a boring background, even though there are probably a lot of boring dead trees to be found in reality.
"The machine is genuinely colour blind. It can't be sexist, or racist, or any other kind of ist because it's not aware of any such concepts."
It doesn't have to be aware to have that outcome, in either direction. They're feeding in unverified junk from the internet, which is made up of all of us. A number of us are prejudiced in some direction, and if we just let the bot show us what it saw, it could easily reproduce that. It doesn't matter that it's showing that prejudice because it ingested some input from someone online rather than being programmed deliberately to do so, at least it doesn't matter to anyone affected by what it does. This experience demonstrates it as clearly as the opposite: it can be trained or prompted in such a way that it produces incorrect information without that being the intended result.
Sounds good, as long as you're willing to be the kind of person I've seen as a programmer who tells me not to tell you why I think this is a bad idea, I build exactly what you said to, and then the game of who to blame for it not working right starts up. Programmers can easily misunderstand what is needed, but usually, it isn't because they got a clear answer and rejected it. Usually, it's because their assumption on what a vague spec means was wrong. That is why, whenever possible, I spend a lot of my time making sure that I understand what you need and you understand what I can do and what I'm currently planning to do. Skipping or rushing that process usually produces code that neither of us wants to see.
Only when I've debugged it for weeks and have started to come to the conclusion that, maybe this time, it's not my fault. It turns out that it usually still is my fault. The reason I can do this is that the compiler gives me the same result each time I compile the same code, so it is possible to change something, see if it helped, and change it back if it did not.
And sometimes, it is the fault of the tools. While I tend not to find many compiler bugs, I have found and fixed bugs in standard libraries before. It starts the same way with me assuming that my code's wrong somehow and I can't see it, and I start to read the library code to see what it is doing with my input because I can't see it. Sometimes, I notice that it's the one doing the wrong thing.
Not at all. Just as with cryptocurrency before it, Nvidia isn't invested in what the process makes. They're not the gold miners. They're the shovel sellers. Shovel sellers often do well. OpenAI and other LLM farms connected to big tech will be okay. It's the smaller AI startups that will take a hammering, and they may not even need the bubble to burst to have that happen.
Throughout the development of tools, we've reduced the number of people required to write any given program, and by doing that, increased the number of people who want programs written. Yes, it will probably happen again, whether LLMs are good enough to do so I'm less sure, but something will eventually start to automate the basic programs that some people have to write. I don't think the people having to write them will mind too much, both because having to write that kind of thing is really boring and because there will be more things that need writing. This is a problem for someone who wants to learn one tool and keep using it forever, but that hasn't worked for programmers for any of the previous years and they should know that it's not going to work now either.
That's just syntax, though. There are languages that can express that sort in fewer statements, requiring less knowledge of the internals, and obtain the same result. A search like that in SQL looks a lot like your sentence, except that you'd use specific column names instead of things like "the vendor" and "part number". Something could take a sentence that looks natural and translate that into the sort you want. That is just a compiler that uses a language that's closer to English. I've seen a few of those, though I tend not to like them.
What you're describing is a bit unclear. It's easy to imagine a program capable of understanding a broad concept: "I want a program to translate text from one human language to another supporting at least English and Swahili", going out and building such a thing. It's not easy to imagine how you'd actually obtain it, but that's what the hype around AI appears to be doing. Meanwhile, your version appears to involve someone being much more specific about how the job gets done, but doesn't end up bringing back all the restrictions that being that specific tends to entail. You're clearly not saying that you can have a one-sentence spec, but I'm having trouble figuring out how your spec and a program differ except that more of your spec looks like English.
Yes, that is a computer language, just one that has a lot of syntax options. There is a lot of that in languages already.
English text: "if 8 in list_of_numbers then"
Mathematical formula: Well, duh.
Flowcharts: I don't like those languages much and they're most often used to try to teach programming, but they exist.
Laws: You can drop one in. It will be your fault when the AI doesn't properly encode legalese into logic appropriate to your situation, since even a programmer nowadays requires a lawyer or two to do that correctly.
I get what you're trying to say, but a program that reads a certain form of input data and produces a program is just a compiler. The syntax it understands can become more and more complex, but if it's deterministically taking an input text and producing an output, it's parsing a language. In order to be deterministic, that language will have to have certain limitations on it as well. I'm not sure the distinction is as clear as you think it is.
It doesn't matter how many hoops I jump through. If I find that a certain set of queries to OpenAI's servers gives me the source code to GPT, that does not make it my source code. I had to jump through a lot of hoops to retrieve it, but my willingness to take that action does not change the legal situation I'm in. The law considers the results and the intent, not the bit in the middle. Taking the code or the articles without permission is illegal. The only reason why getting the quotations is necessary is that OpenAI would, without them, be willing to lie that they were not in the training data. This action does not create or cancel the crime. It only provides evidence of it.
They need the regurgitation demonstration to prove it unequivocally to the court. OpenAI and its adherents have frequently come up with arguments as to why their output is the result of convergent evolution, it's just like any brain, etc. That is more difficult to argue when the quotations are direct. The issue is as important whether the regurgitation is total or partial, but that may not be as obvious to everyone, so showing that it can be made to quote the whole thing demonstrates the violation more clearly.
In principle, this is true, because there are different security requirements based on what the likely consequences of compromise are. However, I still have to disagree with you because it's the argument that people always bring out. What they say is "the potential damage if anyone breaks into this is so small that it's not worth building it more securely". What they mean is "building it more securely would take more time or money and I don't want to".
The world contains a lot of people who assume that security isn't their problem and rationalize it using the same arguments you make. I work in security, so this bothers me a lot. It reminds me of a colleague I had when I was just starting my career who approached all questions of security using the following criteria: if a nontechnical person can break into it, then we should secure it (E.G. if the system logged in automatically or if the password was password, I could do something). Otherwise, since it wouldn't stand up to a concerted attack by China (it was always China), why bother to do more? Something as basic as encrypting drives on laptops was dismissed as unnecessary because his concept of a nontechnical attacker wouldn't know how to get around that, so as long as the laptop asked for a password before logging in, it should be good enough. I don't think you have the same mindset as that guy. I think the argument you make makes it too easy for that kind of guy to rationalize his conclusion.
True, but after you've done that, you need to log in and reconfigure it. I don't envy the customers who don't know that the experience of being told by a chatbot mangled instructions of configuring an ISP-supplied modem. I pity their technical friend or relative even more. Having recently spent hours on the phone to customer service for a friend's ISP to get them to allow the modem to talk to their network, even though I could see that it had already managed to communicate and be issued an DHCP lease, I know the chatbot won't be doing that for anyone.
"What, you don't think computers should act the way *we* would like them to,"
Not really. Basically, if you want the key to mean advance to a new line but don't actually add an explicit line break, then write your own program to do it. I have no problem if you choose to do that. If you choose to use the tool that's already there, spend a little time learning how it works instead of demanding that it do something there is a perfectly good reason for it not to do. As a programmer, I'm happy to explain how it works. I'm happy to explain why it does that. I'm happy to consider why an option to do things differently could be useful in a situation I hadn't considered. I am not happy adding lots of option switches for bad reasons then trying to work out all the other changes needed to accommodate a preference that the user could get around by spending a few minutes learning how it works.
An email summary that gets sent to everyone? You're stuck in the first decade of this century (lucky for you). Nowadays Grandad, we use bespoke web platforms that store every tiny detail of a project in its own task page which has about a hundred buttons and twenty text boxes for different details, none of which has anything in them because filling in a big description of tasks that will take you an hour to complete is pointless. You're supposed to open, close, and update them yourself, but nobody does it because, even if you do, there will be a meeting where someone goes through every one on the board, one by one, and takes a few minutes to hit the resolve button. They're designed to look like they could possibly be useful if they were used differently, but I'm not sure whether that's ever happened.
True, but it is where the legal ambiguity starts to come in and, when it does, the average consumer starts to worry about proving what could turn out to be a simple case. For example, in this case with the broken phone, the charging port was working when it was new, and now it's not. I don't know exactly why as it wasn't mine, but it probably wasn't someone pounding it with a hammer. I could try to suggest it's manufacturer's poor workmanship and they can try to prove that it was caused by user negligence. The typical user looks at all this and decides that, since this was a cheap phone, it will take so much effort that, even if they win, they've probably spent more than paying for a repair or replacement would cost and they're not confident they would win anyway. That is how companies can use disclaimers, even invalid ones, to blunt consequences. The only way around this is if the disclaimer is ruled invalid and they are forced to remove it entirely. If some company can find a wording that the court accepts, everyone will use something similar.
Of course you do if you can hide the disclaimer and have no humans. If they find some set of conditions under which a court allows them to lie via chatbot and not have any consequences, they can use those conditions. Anyone wanting information may try to call someone, but if they only give them the chatbot, then many customers will use it because it's the only option. The customers would be annoyed and some might try not to buy from them, but that doesn't seem to have stopped a lot of companies today who have the bare minimum of customer service.
I agree with you there. If they did it, it would have to be done with some other type of technology, and it's unlikely to be able to do it with perfect accuracy. However, I still think that one of these AI companies might eventually realize that they need it and try to build such a thing. They may fail to accomplish it, but for the moment, nobody is even trying. This is assuming that enough people rely on incorrect chatbot answers and suffer the consequences, so I'm hoping that rulings like this continue to happen. If users manage to find a way not to suffer when they use a chatbot's answer to screw over another, my assumptions could prove wrong.
It depends if a court accepts it. If they basically say that the chatbot can say whatever it likes and they don't have to honor any of it, then it can still look like it's providing customer service and leave every user with the results.
Such things are not uncommon already. I was reminded of this recently when I was asked to repair a phone with a broken charging port which had been purchased only months ago. The warranty attached to it had so many different reasons why something wouldn't be covered that, as far as I could tell, the damage was not covered. The customer service person said that it was not covered, but not why. What I can't figure out is what kind of damage, other than maybe the phone being broken before anyone touched it, would have been covered by the warranty. Yet a customer buying the device would think that they had some kind of protection anyway because the warranty existed, and surely they wouldn't have a document if it meant nothing.
The average user will probably never see the disclaimer and assume that, when the chatbot the company chose to put there gives them some information, it is valid. It is possible that a court will overturn that and invalidate the disclaimer, the same way that I could probably have tried to challenge the warranty, but most users will not try because there is a good chance that it won't work and they'll end up wasting time and money in the attempt.
Admittedly, its first answer could be due to the vague question. "the VAT paid on £500" could mean £500 total expenditure including tax or £500 before tax. If I were asked the question, I'd probably ask for clarification. If I couldn't have it, I'd use context to guess which was wanted, for example using a before tax amount if the person asking was a seller and after tax if it was a buyer, but that isn't a guarantee of anything.
I wouldn't be surprised that OpenAI will eventually try. They've made it so far by having a bot that can print coherent sentences, and they've succeeded at convincing some companies to pay for it. I think that ascendance will eventually break when the inaccurate results become bad enough. If they can see this coming, they might start focusing on getting some accuracy out of it, not just hiding their training data. The money they've made from sales so far and from massive investments from Microsoft should allow them to contribute resources to the attempt. I don't know how easy that will be, and they may try to get it and fail, but I do think they or someone like them will try.
People can take some levels and forms of unreliability. For example, when they turn on their navigation systems, they know there's a chance it might send them the wrong way, but they also know that most of the time, it is correct, and often if it is not, they can use their eyes to detect it and their physical control of their vehicle to prevent it from causing problems to them, the odd story about a person deciding to drive directly into a lake excepted. If the navigation system made errors like that on a significantly larger set of trips, people would be more cautious about using it, as they are in areas where map data has not been updated. If those errors were not immediately detectable and frequently led to wasted time or risked lives, it would be dangerous for users to rely on it.
GPT's reliability and accuracy is along those lines, since, as you know, it is frequently wrong and it doesn't make that obvious unless you already know it. People who trust it are making a mistake, and people who use it in such a way that others are depending on GPT-derived unreliable data are harming others. I have been clear about this, and others have stated their opinions. Your repeated assertion that we must be saying these things out of fear is beginning to confirm my original assumption that you know what we think, cannot respond to it except by trying to understate the inaccuracy of GPT, and are resorting to making up a different reason and putting it in our mouths. I can only tell you that, if you sincerely think I and many others here are actually fearful of GPT's effect on our jobs, you are profoundly mistaken.
A bad workman blames his tools because he uses crap tools. A good workman does not need to blame his tools because he chose the right ones. GPT is not the right tool, and using it indicates that something is going wrong in the researcher, and if they're not exceedingly careful, the research. Similarly, a researcher who asks someone who has no reason to be trusted as an authority is also doing research wrong, even if they put little chunks into a search engine and find a page that says something that looks similar. They actually have to understand what they're reporting on to do research correctly.
What you appear not to be taking into account is that GPT changes a lot. It's not just because the algorithm has a lot of random elements, but that there is a lot of premade prompt material which is added to your prompt before you start, OpenAI changes it frequently, and you can't see what it is. What one prompt produces today may not be generated, even similarly, by the same prompt a day from now.
This "what is a computer" example is a stark demonstration of the effect. It did actually happen. OpenAI verified that it happened. This is not supposition. It's working now because people posted that and other examples where it was clear that someone had flubbed something very important, making a lot of prompts produce gibberish. This works for other cases as well. If I post a prompt which returns completely made up information and you run the same prompt some time later, you may get the same inaccurate information, an I don't know response, completely different inaccurate information, or even correct information. There's really no way to tell. Asking people to post prompts then pointing out that you can get something different suggests that you might not know what is happening when you do so.
Since we were discussing it in a thread above, this inconsistency is one of the reasons why I have a problem with the use of GPT when accuracy is important and one of the reasons I thought you might be trolling. Even getting a useful answer once doesn't guarantee that you can get a useful answer later with the same prompt.
I will try to explain, assuming that you really want to know our views and aren't just trying to troll. First, I need to clear up some problems with what you asked. We aren't saying what we do because we're afraid of GPT, at least I am not. I do not fear it, and I'm not worried about losing my job to it. Call that overconfidence if you like, but that is not a consideration for me.
My problem with GPT and models like it is that it is so frequently wrong that it is bad to use it for situations where accuracy is important. Most of the time, accuracy is very important. I have had similar problems with many things before. For example, I know some people who, when they're feeling a bit sick, will try to use Google to diagnose themselves. Their arm aches because they did more work with it than they usually do, but they're now worried about cancer. That is something I'd rather they not do. It's not that any particular page is lying to them, but that it's not the correct tool for what they want done. GPT is fine if you want something fictional generated, and it may work if you want a virtual conversation, but it is not a good tool for research.
I do not see it as similar to Wikipedia at all. Wikipedia gave you lots of details with the possibility that some of them were wrong or that some important information may be missing. GPT gives you summaries with the chance that everything it's saying is rubbish and no way to get more information. I have tried it numerous times and, whenever I ask for details that aren't commonly found, it either fails to find them when a simple search could turn them up or it makes them up. Someone who relies only on Wikipedia is likely missing something, whereas someone relying only on GPT is probably going to do something catastrophically wrong sooner or later.
"Is "insure" vs "ensure" a UK vs US ( two countries divided by a common language) thing?"
According to my experience with the language and a quick check of a couple dictionaries, no. The spelling and definitions are the same. It was probably just a typo.
I think it's likely a combination between someone being in an area where other providers' signal is weak and having phones that lack some of the bands that other providers would be using in the area. If someone on a different provider would be using a band that wasn't included on the device because that one was built for AT&T networks, they would be unable to use that one to connect to emergency services. This would indicate why it was just some customers, not everybody in the city, who can't connect. It would be easier to demonstrate if we checked the devices that didn't work. Most flagship devices have a lot of bands, but some cheap ones, especially the cheap ones sold under the providers' brands, go for the basic minimum.
I agree with you on getting explanations for why you said what you said, although that's far from perfect, but I can't agree with this bit:
"Humans are far more debuggable right now because we have 50,000 years of learning to understand each other."
And despite that, we can almost never debug something even when we've decided it's going wrong. I can't tell you why interactions between humans go wrong except in broad terms, and I can't go in and fix them. I can't explain why there are dictators who prefer to kill people even when they don't have to, nor why people support them, but they do. Those are some pretty big bugs in my opinion, but we have no chance of fixing them any time soon. It works at the micro level as well. There are lots of mental disorders that don't have an established debug process. The best we can get in many cases is a mitigation, and those often fail. We have psychologists, neuroscientists, and to some degree geneticists working on that, but those efforts will take a long time to improve.
An LLM, despite its general uselessness, is easier to control. If OpenAI doesn't want it to quote copyrighted material, they can significantly limit that behavior, although they can't eliminate it entirely. If you want to avoid thinking something, it will take either a significant amount of effort or it could prove impossible. The LLM is easier to debug.
It might be if the job might require you to go somewhere. If, for example, I was the person they'd call if they wanted something physical changed, then I'd have a job requirement to go where that thing was in the time they expect. If my job involved going to in-person meetings in a certain location, they might have a similar requirement. Which house I live in is none of their business, but the general location can be. I don't know whether these considerations apply to the commenter above, but they might.
That's not really what I intended. That's RAM caching, which is already available and works fine although if you don't have a lot of RAM, the caches will be rebuilt frequently. I was trying to talk about the initialization process of a typical program today, which involves copying its machine code into memory, then creating a lot of runtime data. Much of that data is loaded verbatim from files on disk, but some of it is calculated at startup. Theoretically, you could precalculate any static data and store that in the binary instead of computing it when the binary starts, which would decrease loading times a little. However, it's not going to reduce it too much and it is already an option.
They can do that, but it won't change how mass storage is used. For example, when a file needs to be expanded, it's still going to be fragmented rather than treated as a big string in memory, and to manage that, they'll still need to track which areas of mass storage can be written to across programs. And we've reinvented the filesystem.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
