Comments by "Lepi Doptera" (@lepidoptera9337) on "Computerphile" channel.

  1. They shouldn't teach you OOP in high school to begin with. Chances are that they are screwing up your mind with all the wrong ideas about what OOP does.
    4
  2. The problem with that is that you are NOT doing an actual matrix multiplication that way because the math that uses the matrix multiplication describes an infinite repetition of the quantum computer and not one copy of it.
    3
  3. And if function doesn't do anything additionally, then it's useless. So by your definition all programs that do something useful are bad design. ;-)
    2
  4. It won't. Neither will those code generators.
    2
  5.  @nfiu  Transformers, just like every other tidbit of current AI research are more hype than anything. At the end of the day it turns out that it's not all that simple to replace a human mind. But let's say you do achieve that, then you are still up against the obvious: the halt problem and a slew of similar results which are all related to Goedel's theorems and which make it essentially impossible to prove the correctness of software. Neither man nor machine can beat mathematics.
    2
  6. A pure function is useless.
    2
  7. If a program doesn't have a side effect, then it's useless.
    2
  8. And that way you have never written anything resembling complex software, either. Pure functions are basically just breaking a long piece of code into shorter pieces of code. They are doing absolutely nothing for you in terms of actual problem solving. Having said that, if you have used even one IO function in your life, then your statement that you have only written pure functions is not even true.
    2
  9.  @atlantic_love  All of this can be done easily and without any confusion with libraries.
    2
  10.  @atlantic_love  Replace "sorting" with "any algorithm of your choice". How likely is it that such an algorithm applies universally to more than a few classes? Very small in applications that I have seen during my career. That is one of the problems with OOP. It makes you write a lot of fluff but doesn't give much in return.
    2
  11. Because this is the internet? ;-)
    2
  12. You can't do it in the US, either. You can't send out HR information to non-certified companies.
    2
  13. There is no catch unless you are an investor in one of these companies. You can download the model parameters and load them into your inference engine. Then you can use the model. Whether it will do something useful for you is a different matter.
    2
  14. No, the actual point was to talk about computability and complexity and it does that fine. What it doesn't do is to give you recipes for programming real computers for real applications. This is applied mathematics, not actual engineering.
    1
  15. A classical computer does. That's why we feel comfortable programming them. This is not true for a quantum computer and that's why basically nobody knows how to program them at all.
    1
  16. Yes, and you have never seen a purely functional program that does something practical in your entire life. Why? Because practically useful programs need state. They can not be written as pure functions.
    1
  17.  @alpo789  I never understood why people think that "root" is safe. For 99% of users it's more likely that the hacker knows how to become root before the user does.
    1
  18. They are simply partying like it's 1999, again. Meta has probably spent more money on the completely useless Metaverse, so far, than the cost of AI has been. It's just endless advertising dollars and investor dollars being burned, that's all.
    1
  19. "Build a website.". Is that enough of an intention description to build an empty HTML document? Sure. How do you get from that to $100 billion in revenue like Facebook? See the problem? :-)
    1
  20. An NDA is a legal document that defines a "reasonableness" standard for the safekeeping of trade secrets. It prevents both sides from bringing nonsense lawsuits. If you are ever exposed to somebody's trade secret without having a written NDA in place, be very careful. It might backfire if you are dealing with a possessive personality. With an NDA all you have to do is to keep their trade secrets as safe as you would your own, i.e. they can't require you to pay damages for accidental leaks if you abide by the low standards of the document, which are usually trivial. If you are used to keeping your own trade secrets in a file folder in a locked office and you have employment agreements that require your employees to keep their knowledge about your company and its operations to themselves, then you are done implementing security measures for your partner as well. They can't sue you for not keeping their documents in a safe inside a vault inside a military installation with double fences and armed guard towers. ;-)
    1
  21. Graham Hutton wasn't even born when the early papers on parsing were written. Is this dude even real?????
    1
  22. It is not the academics that are patenting inventions but their universities. As a researcher you do not own your inventions. They belong to your university. That's in your contract.
    1
  23. The mere idea that security exists in this space is a ridiculous misunderstanding of physics. A radio transmitter can always be located simply by the fact that it has to produce an energy flow that is above the noise background of the environment. No matter the protocol, it is always possible to detect the source of the transmission. For a criminal the detection of a police transmitter close to his physical location would usually be enough to seize the criminal activity. It is complete overkill to differentiate between "harmful" and "harmless" police presence for most such activities. That's why the police usually does not care about being listened to.
    1
  24.  @StefanReich  Which is just as bad an idea as this. A user machine should NEVER allow root access.
    1
  25.  @tablettablete186  sudo is a naive workaround for the failed "root" concept. The average user should NEVER have to use it. Look, all of this was invented at a time when computer security was not a problem. Today it is. Nobody was doing their bank transactions on a multi-user mainframe in the 1960s and early 1970s. Except for the bank, that is... and they could physically secure their hardware and the only possible "hackers" were their own employees who had hardware access. That made these simple minded concepts somewhat workable. They are not workable today.
    1
  26. Curiously, it's the most secure thing you can do, if you use one time pad ciphers. And honestly... why would you use anything else in a day and age of 4Tbyte SSD drives? One drive is enough for years of voice communications. ;-)
    1
  27. I don't know what people are going in about with regards to prompt injection. I did not have ANY success in using context to get AI to "unlearn" its training data. Once an answer contains nonsense any and all repetitions of the same question in different forms will return that same nonsense in one way or another. Current AI models simply do not learn on the fly.
    1
  28.  @0LoneTech  Be that as it may one can not make LLMs unlearn nonsense, which would be absolutely necessary to make them useful in non-trivial domains.
    1
  29. I don't understand the point you are making here... proprietary technology (copyrighted, trademarked, patented or even just kept as a trade secret) is just like a "top secret" marker on government files: it's a declaration of legal ownership. It's not a matter of security. Stamping "Top Secret" on a document doesn't make it safer. It does, however, guarantee that the government can prosecute unlawful use.
    1
  30. There are no identifiable physical states there, at all. A physical state is like the outcome of a dice throw, i.e. like "1" or "3". When the ensemble of a quantum system is in superposition that's like having an infinite number of dice throws going on at the same time. Are those moving dice in an identifiable state in their outcome space (1-6)? No, of course not. Neither are the quantum systems.
    1
  31. A language doesn't solve problems. What a crock. :-)
    1
  32. A measurement does NOT make the "probability" of an outcome 100%. That is absolute nonsense and folks who are speaking on a math channel should know better than this. A measurement simply produces one physical outcome. If we tabulate many such measurements, then we get frequencies and if we then imagine that we can take an infinite number of measurements, then we get probabilities through the law of large numbers. The individual measurement, however, does NOT change such probabilities at all. It has an infinitely small influence on the probabilities by definition.
    1
  33. Why in the world would you use "Pong" as an example?
    1
  34. Functional programming is for those who didn't pay attention in CS 101 when computer architecture was discussed and who want their compilers to auto-magically solve all the problems that are the result of real computer architecture.
    1
  35. No, it's not. It's not even close. Neural networks are perfectly classical systems.
    1
  36. No, you can't do that. Quantum mechanics does not work like any of the digital systems that you may be familiar with.
    1
  37.  @Wabbajack-kj2kg  The entire purpose of your computer is to produce side effects. Functional programming simply misidentifies the problem. The problem is not to avoid side effects and state. The problem is how to manage side effects and state safely. The latter requires an understanding of what causes state modifications to be unsafe (non-atomic operations, for instance) and most of those real problems can not be solved with a programming language paradigm. They are not a matter of grammar at all.
    1
  38. It kills it. ;-)
    1
  39. As soon as that happens you won't have a job, kid. Be careful what you wish for. ;-)
    1
  40. Not much. The returns on investment are, at most, logarithmic. These models will get better over time but not because of size.
    1
  41. Trivial answer: no. You can't do any of that because you can't even predict the output of a Turing machine algorithm, which is perfectly deterministic. These systems aren't even that.
    1
  42.  @atlantic_love  Any language. A universal sort algorithm is nice, academically speaking... but in practice it doesn't matter. How many hundred different types are you sorting in a single application? And even if you want to sort different types with the same algorithm, then what does it really take? A few function pointers in C. That's it.
    1
  43. Neither of which can do anything useful, either. ;-)
    1
  44. There must be about a million JavaScript libraries which can do that for you. So what? So nothing. This is like me giving you thirty tubes of oil paint, a canvas and three brushes and telling you to paint the Mona Lisa.
    1
  45. 0.01% of all users will do that. ;-)
    1
  46. Yes, and none of it will lead to a stable business model. Nobody is willing to pay for AI tools. Look at search... which is enormously useful. Nobody has ever been willing to pay for search, either.
    1
  47. So the lambda calculus can only encode pure functions. Real programs with state: hold my beer.
    1
  48. What stops me from writing acyclic code that will take longer than the lifetime of the universe to terminate? The entire idea sounds naive to me. Unless they are setting a low (1 minute) limit on top, without the ability to restart... and even then one can probably still do some rather malicious things with it.
    1
  49. That's not a sign of manhood. It's a sign that he has a clinical case of needing to be different. ;-)
    1
  50.  @sebastiangudino9377  It doesn't come with Windows, child. Why would it? Windows has specialized IDEs of all sorts for all kinds of R&D. I am currently using Eclipse with an embedded microcontroller environment. The hardware debugger is fully integrated with the code editor and I can step through my program line by line. I could never do that with vi. OK, now you got your two minutes of attention. I hope your basement has gotten a little warmer. ;-)
    1