• kescusay@lemmy.world
    link
    fedilink
    English
    arrow-up
    69
    ·
    edit-2
    9 months ago

    Copilot isn’t actually bad for developers, it’s just that you need to be careful with it and recognize its limitations.

    Writing a bunch of REST endpoints for an API and need to implement all the typical http verbs, and you already have all the matching methods for reading, updating, and deleting values in a complex SQL database for each endpoint to call? Copilot can turn a ten minute chore into a ten second one. Very handy.

    Writing those complex SQL methods in the first place? Yeah… Copilot will probably make a ton of mistakes and its work will need to be triple-checked. You’ll save time just doing it yourself if you know how. (And if you don’t, you have no business calling yourself a developer.)

    Copilot is best for easy boilerplate and repetitive code. Problems arise as soon as you ask it to get “creative.”

    • TimeSquirrel@kbin.social
      link
      fedilink
      arrow-up
      14
      ·
      edit-2
      9 months ago

      One time I decided for shits and giggles to just keep pushing tab and see where it went. It didn’t take long for it to enter a useless recursive loop, hallucinating a new iteration of the same thing on each line.

      It definitely isn’t gonna magically think up new algorithms for you. I don’t know what everybody is scared of. It ain’t even gonna replace my kid programming on Scratch.

      • kelvie@lemmy.ca
        link
        fedilink
        English
        arrow-up
        3
        ·
        9 months ago

        I mean didn’t we all do this when phones started autocompleting sentences like a decade ago? (Or however long it was, time perception is fickle)

    • lemmyvore@feddit.nl
      link
      fedilink
      English
      arrow-up
      13
      ·
      9 months ago

      Copilot isn’t actually bad for developers, it’s just that you need to be careful with it and recognize its limitations.

      Is it me or is this a weird statement for what’s supposed to be an exact science?

      Imagine working in construction and using a level and you’re told “it’s not that it’s a bad level, you just gotta be careful with it”.

      How much margin for error should we allow for getting our code right? Is it now acceptable if we only get 80% right?

      • kescusay@lemmy.world
        link
        fedilink
        English
        arrow-up
        8
        ·
        9 months ago

        It’s more like you get some kind of weird construction multitool that promises to be a level, a drill, a hammer, and a dozen other things, and it turns out to be a really good, innovative, and helpful level… and a really bad everything else.

      • pezhore@lemmy.ml
        link
        fedilink
        English
        arrow-up
        5
        ·
        9 months ago

        I use copilot a bit for my work - and I treat it like copy-paste from StackOverflow - sure that codeat look right, but you’ve gotta double check it and test it a few times before you commit and push.

      • birbs@lemmy.world
        link
        fedilink
        English
        arrow-up
        4
        ·
        9 months ago

        As a software developer I promise you that software development is very much not an exact science.

        Programs are complex and there are so many different ways of achieving the same thing that all code has problems and gets a bit messy in places. You can test, but it’s not easy to ensure that everything works the way it should.

        The best code you’re going to get will probably be in the space industry, but even that will have bugs. The best you can do is make the code robust even when bugs make things go wrong.

        In many cases copilot will do just as well as a junior developer. It’s very good at repetitive tasks and filling gaps in your existing code.

    • friend_of_satan@lemmy.world
      link
      fedilink
      English
      arrow-up
      4
      ·
      edit-2
      9 months ago

      Always ask it to write tests for the code it generates. Of course, then you have to validate that the code works AND that the tests work.

      • anotherandrew@lemmy.mixdown.ca
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        9 months ago

        Or just write the damn thing yourself and save a bunch of headaches and wondering if you got the tests right it if there’s some screwy corner case lurking because of its implementation.

  • ShittyBeatlesFCPres@lemmy.world
    link
    fedilink
    English
    arrow-up
    46
    ·
    edit-2
    9 months ago

    It’s funny that Microsoft fucked up the branding so bad that half of these comments are about GitHub Copilot — a specialized, useful tool for experienced developers to speed up rote tasks — and the other half about their whole “Shoehorn AI into everything.” strategy. And Bing Chat is also now Copilot, apparently? Is Cortana now Windows Copilot? Is there an Xbox Copilot that plays Starfield for you?

  • Eager Eagle@lemmy.world
    link
    fedilink
    English
    arrow-up
    45
    ·
    9 months ago

    Similarly, other users told the outlet that the AI hallucinated wrong answers or miscalculated spreadsheets. AI experts, including The Wharton School professor, Ethan Mollick accused Copilot of making bizarre suggestions for weekend meetings.

    It seems these users never used GPT

  • FMT99@lemmy.world
    link
    fedilink
    English
    arrow-up
    31
    ·
    9 months ago

    … In contrast, Microsoft’s corporate VP, Jared Spataro, told us that users are finding immediate value in Copilot

    haha no really?

  • Rexios@lemm.ee
    link
    fedilink
    English
    arrow-up
    24
    ·
    9 months ago

    So people are surprised that GPT-4 is performing as well as GPT-4 always has?

  • BrightCandle@lemmy.world
    link
    fedilink
    English
    arrow-up
    13
    ·
    9 months ago

    As far as I am concerned Copilot is a giant theft of open source code and breaches the license. I expect in the future a lot of repositories will be used to poison these AI agents just as is happening with images. The agents will get better but the quality of what they produce will also be poisoned and get worse precisely due to the theft.

    • barsoap@lemm.ee
      link
      fedilink
      English
      arrow-up
      7
      ·
      edit-2
      9 months ago

      Poisoning code should be ludicrously easy: They crawl pretty much everything and a random AST walk looks suspiciously like real code while it’s the equivalent of showing an image generation model noise. Or maybe better: Mondrians that are indistinguishable from Vermeers. (I hope I didn’t offend anyone by calling Mondrian abstract nonsense but it is abstract nonsense).

      I don’t think copilot will hold out for long anyway, the novelty is wearing off and even inexperienced programmers are beginning to see that it helps you write code faster that shouldn’t have been written in the first place. Code is like 90% maintenance and excessive boilerplate doesn’t make it easier.

      OTOH please don’t let that “Let’s scam artists by selling them snake oil that if it wasn’t trivial to circumvent would break naturally within a week” guy fool you. On the actually interesting side of poisoning attacks, people have made cars hallucinate radar blips I bet a couple of companies are getting quite tough questions from regulators right now.

  • 7rokhym@lemmy.ca
    link
    fedilink
    English
    arrow-up
    10
    ·
    edit-2
    9 months ago

    Yeah it’s a classic case of Microsoft marketing. So far I found the office integration to be the least useful and most over hyped in marketing. However what it is good at is actually helpful. Join a meeting late it already has an update for what’s happened on the meeting so far and it’s really good for summarizing a meeting especially a key topics and a summary of action items. Tedious tasks like taking data copied from a PDF file and reformatting it correctly in CSV. And my favorite is making custom graphics based on a specific colour palette, though most images are really good for entertainment, demos and samples, but not production quality for final products. Weird results include creepy human images just don’t look right in a disturbing zombie-like way.

  • nyakojiru@lemmy.dbzer0.com
    link
    fedilink
    English
    arrow-up
    9
    ·
    9 months ago

    It’s ChatGPT Open AI, but as Microsoft LoVES to do things, bloated as those old time web browsers infested of tool bars

    • Hotzilla@sopuli.xyz
      link
      fedilink
      English
      arrow-up
      1
      ·
      9 months ago

      It is multiple things, in chat it is just GPT4-turbo, with support to request data through Microsoft Graph API. So basically it can do API searches to data you have access.

      In Excel, it can perform things like formulas, with context of your excel sheet. This is actually quite useful.

      But still bit disappointed, the system prompt of copilot is in my opinion not as good as open AI’s.

  • esc27@lemmy.world
    link
    fedilink
    English
    arrow-up
    8
    ·
    9 months ago

    I have not had a chance to test the office integrations (and the $30 price tag may keep it that way,) but I’ve been testing the web chat version, and once toggled to “precise” mode I find it close to chatgpt. Just a little more prone to lapse into acting like a bing chat app and very limited in conversation length…

    I think the trick is to treat it like a junior assistant or maybe an intern who might make mistakes and not as a seasoned, experienced employee who always puts out perfect work.

    • misanthropy@lemm.ee
      link
      fedilink
      English
      arrow-up
      8
      ·
      edit-2
      9 months ago

      $30? Idk how it is in the consumer space, but in the Enterprise world the initial opening for copilot required 300 seats of E3 or better, and then a purchase of 300 seats of copilot 365 at $30 each. They were supposed to drop the 300 copilot 365 seat req q1 but I’m not sure if they did.

      3030012=108,000 per year for copilot. E3 is 36 a month, 30012=129,600.

      Big money.

      • esc27@lemmy.world
        link
        fedilink
        English
        arrow-up
        5
        ·
        9 months ago

        I’m told they dropped the seat requirements. Yeah $30 is the business rate. No discounts for education. I suspect the backend costs on this are still rather high but should improve once hardware catches up to demand.

        • misanthropy@lemm.ee
          link
          fedilink
          English
          arrow-up
          1
          ·
          8 months ago

          Ah. Well as far as I was told two months back, they didn’t plan on dropping the E3/E5*300 seat requirements, only the 300 seat of copilot requirements. But I haven’t kept on top of it, all my companies clients are too small fish to drop 100-200k a year on unproven tech

          From what I’m told, Ms was basically running things at a loss when they first opened the doors.

  • SrTobi@feddit.de
    link
    fedilink
    English
    arrow-up
    5
    ·
    9 months ago

    This is certainly not about finding fools who not only are beta testers but also provide training data… And not only for free, but they pay also for this… Win win win for M$

  • whoelectroplateuntil@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    4
    ·
    9 months ago

    My favorite are the gigantic companies falling all over themselves to buy this for their workforces before it’s even clear what value a bot that can’t reliably do arithmetic is going to add to Excel

    • EatATaco@lemm.ee
      link
      fedilink
      English
      arrow-up
      1
      ·
      9 months ago

      According to the article, dow says they’ve seen significant productivity gains and are rolling it out to half the company.