Some middle-aged guy on the Internet. Seen a lot of it, occasionally regurgitating it, trying to be amusing and informative.

Lurked Digg until v4. Commented on Reddit (same username) until it went full Musk.

Was on kbin.social (dying/dead) and kbin.run (mysteriously vanished). Now here on fedia.io.

Really hoping he hasn’t brought the jinx with him.

Other Adjectives: Neurodivergent; Nerd; Broken; British; Ally; Leftish

  • 0 Posts
  • 138 Comments
Joined 6 months ago
cake
Cake day: August 13th, 2024

help-circle

  • It’s kind of easy to forget about or ignore any experience they might have if they’re asking questions like that. Sure, maybe it was a brain fart from a panicked intern who’s having orders barked at them from a powerful individual that they want to impress, but that doesn’t make it any better, does it?


  • not sure what they’re trying to do here

    Maximise profits and minimise losses. My guess is that someone important at Microsoft thinks that this will do just that, and if not that, will make them, personally, a lot of money. That person has no-one who will dare challenge their authority and so we go down this road.

    They (that individual or Microsoft as a whole) almost certainly have a stake in the companies that provide newer hardware, and if they didn’t before this decision, they will have by now.

    It theoretically makes Micosoft’s job easier too. A huge chunk of backwards compatibility maintenance goes out of the window, if you’ll pardon the pun.

    “Oh you have 5 year old hardware? We don’t support that.”

    Sounds fairly similar to Apple’s business model if you think about it that way.


  • Current AIs often suffer from what’s sometimes called “the strawberry problem” and can be easily confused into doing mathematics incorrectly. There’s also the human element of “if it comes out of a computer it must be right”, forgetting the “garbage in = garbage out” principle.

    The strawberry problem is a colourful name for the fact that LLMs turn sentences into something else internally and can’t then go back and re-examine the input to make checks and comparisons. Thus if you ask an LLM how many 'r’s are in the word “strawberry”, it often gives the wrong answer, unless it has been explicitly told what the right answer is. And now you have no idea what else it has and hasn’t been told is right and wrong.

    As for mathematics, they have to be explicitly programmed to be able to use a proper calculator if they’re to do mathematics correctly. Otherwise you’ll get something that looks good enough to fool someone who doesn’t know any better.

    LLMs are basically lossy compressions of knowledge. At a high level of abstraction, the creation of an LLM is fairly similar to how a raw image is turned into a JPEG*. There’s a necessary, deliberate bottleneck in the creation process that keeps the size down, and that’s going to show up in the output if you look closely.

    Using the output of an LLM is a bit like editing the JPEG rather than the raw image. Some of the things you do will invariably enhance those artefacts.

    For a JPEG that’ll do nothing worse than make an image “deep-fried” or otherwise ugly. Put an LLM in charge of people’s lives and it’ll do the same to them.

    * JPEG has a lossless encoding variant, but that’s not the right analogy here.


  • There used to be an undocumented setting in early versions of MS-DOS that would allow the setting of the command option character to something other than the slash, and if you did that, the slash automatically became the path separator. All you needed was SWITCHAR=- in your CONFIG.SYS and DOS was suddenly very Unix-y.

    It was taken out after a while because, with the feature being undocumented, too many people didn’t know about it and bits of software - especially batch files, would have been reliant on things being “wrong”. The modern support for regular slash in API calls probably doesn’t use any of the old SWITCHAR code, but it is, in some way, the spiritual descendant of that secret feature.

    Here’s an old blog that talks about it: https://learn.microsoft.com/en-gb/archive/blogs/larryosterman/why-is-the-dos-path-character










  • Me: “Oh no, that wtf means it’s gonna be a LiveLeak site isn’t it?”

    platform for general use

    Me: “Ah, so maybe not so bad then.”

    Every topic is welcome.

    Me: “… and yet…”

    Seriously though, I’m all for the Fediverse in all its many forms, and without people like you it wouldn’t exist. Thank you.


  • palordrolap@fedia.iotoProgrammer Humor@programming.devErrors
    link
    fedilink
    arrow-up
    23
    arrow-down
    1
    ·
    15 days ago

    Urge to analyse… rising…

    My first guess would be to take out that semicolon on line 264. JavaScript will often happily take a new line as end of statement if it makes sense to do that, so in theory, that semicolon is not needed. And it might be a Greek question mark your prankster colleague put in your code when you weren’t looking.

    And then I’d be tracing parentheses, curlies, quotes and so on, because that error could be the point the parser gave up trying to make sense of the code rather than where the error actually is.

    And if that didn’t find it, I’d put in a deliberate error at an earlier, known line to see where the parser thinks that error is. If it’s offset by 20 lines, then I know the original error is probably offset by a similar amount.




  • Ah, misleading use of terminology that indicates one thing, but will win in court even if it actually means, or can later be said to mean, another.

    I hope those involved in helping companies win these lawsuits choke on bones from food sold as boneless. Because that won a court case after “boneless” was redefined as a cooking method.

    I don’t want them to choke to death. Just a little lesson, you know?



  • I remember making one of those.

    It had a faux URL bar at the top of both the left and right frame and used a little JavaScript to turn each side into its own functioning browser window. This was long before browser tabs were a mainstream thing. At the time, relatively small 4:3 or 5:4 ratio monitors were the norm, and I couldn’t bear the skinny page rendering at each side, so I gave it up as a failed experiment.

    And yes I did open it inside itself. The loaded pages were even more ridiculously skinny.