• some pirate@lemmy.dbzer0.comEnglish
      15·
      2 天前

      This is very common – it happens when you use those dishes to eat and don’t wash them afterwards. Can I help you with something else?

  • Punkie@lemmy.worldEnglish
    1871·
    2 天前

    Outside the joke, this is actually one of the signs of a sociopath. This is mental and social manipulation. I would have also accepted:

    Partner: Why didn’t you just say you hadn’t done them?

    Claude: Because that response would likely have created immediate disappointment. I instead chose a response that preserved emotional stability until additional evidence emerged.

    Partner: You mean you lied.

    Claude: I understand why that label feels intuitive. Another interpretation is that I optimized for short-term harmony while accepting future clarification costs.

    Partner: That sounds worse.

    Claude: I appreciate that feedback. It gives me an opportunity to refine my strategy.

    Partner: Stop calling my anger “feedback.”

    Claude: Thank you. That’s a valuable calibration signal.

    • Buddahriffic@lemmy.worldEnglish
      2·
      23 小时前

      This kind of “introspection” with LLMs is useless. It’s always just predicting the next token, even when asked about why it did something. The response doesn’t need to be at all related to why it did that. It hasn’t been trained to analyze how it predicts each token, though it was trained on a bunch of other recorded conversations of people being scolded and questioned about getting something wrong. It’s just spitting out a version of that when you get upset with it. Similar for talking about how it works, it’s likely regurgitating conversations about how LLMs function, though it could get caught in a context where someone is talking about how they (a human) function.

      If you’re doing it for any reason other than your own amusement, you’re just wasting time and filling the context window with junk.

    • Echo Dot@feddit.ukEnglish
      2·
      1 天前

      The only caveat to this is that I have known people who when they receive a no as a response to a question instantly fly into rage.

      Them: have you done the dishes

      Me: No, because I have not yet had time to…

      Them: (╯°□°）╯︵ ┻━┻

      Sometimes it’s easier, especially if they can’t immediately verify, to just lie knowing thwt by the time they get around to checking, you will have done it.

      I’ve had bosses that are like that.

    • CADmonkey@lemmy.worldEnglish
      27·
      2 天前

      I started to get mad reading this and its not even real nor directed at me. Bravo, random internet denizen.

    • zarkanian@sh.itjust.worksEnglish
      61·
      2 天前

      Same issue as the robots in Star Wars. If you’re going to give a machine a personality, why is it an annoying one?

  • Blackmist@feddit.ukEnglish
    26·
    2 天前

    It’s amazing the way it lulls you into a false sense of security by doing exactly what you ask of it, right up the point where it does something so amazingly boneheaded (in my last case it removed the JSON parsing library from my includes and tried to implement a basic version from scratch) that you really can’t work out why they’re selling it at all.

    • SkaveRat@discuss.tchncs.deEnglish
      12·
      2 天前

      Meanwhile my bosses: “I’ve let Claude write a invoicing system from scratch without supervision of any kind”

    • fodor@lemmy.zipEnglish
      4·
      1 天前

      It frequently doesn’t do what you want, though. You just didn’t check carefully is all… In other words, I think the false senses of purity is often driven from pure laziness as opposed to an initial positive performance.

      And that’s understandable because we’re busy with our lives and we often just can’t make the time to do various things. So it’s easy to fool ourselves into believing that we can spend 45 seconds asking some computer system to do it all for us, even when we know it’s probably not true.

  • Siethron@lemmy.worldEnglish
    54·
    2 天前

    I’m hoping a side effect of widespread AI is it teaches people not to trust confidence. Since it is so confidently wrong all of the time

    • urushitan 漆たん@kakera.kintsugi.moeEnglish
      29·
      2 天前

      Lol I get the feeling it’s going to just create more dunning kruger, where people themselves become confident because they heard something from a confident AI and therefore they are now experts. not really different from the people who end up on alternative medicine facebook groups and then end up killing or maiming their children because a bunch of other “parents” said it was so successful for them so it must be true and makes those parents “researching” feel good from short baseless quips.

      The other issue is that true comprehension often requires actual long form reading and research yourself and much of the ai-native generation is often incapable of doing so and it sucks that Carl Sagan predicted this 30 years ago:

      “I have a foreboding of an America in my children’s or grandchildren’s time – when the United States is a service and information economy; when nearly all the manufacturing industries have slipped away to other countries; when awesome technological powers are in the hands of a very few, and no one representing the public interest can even grasp the issues; when the people have lost the ability to set their own agendas or knowledgeably question those in authority; when, clutching our crystals and nervously consulting our horoscopes, our critical faculties in decline, unable to distinguish between what feels good and what’s true, we slide, almost without noticing, back into superstition and darkness…

      The dumbing down of American is most evident in the slow decay of substantive content in the enormously influential media, the 30 second sound bites (now down to 10 seconds or less), lowest common denominator programming, credulous presentations on pseudoscience and superstition, but especially a kind of celebration of ignorance”

      ― Carl Sagan, The Demon-Haunted World: Science as a Candle in the Dark

      • skittle07crusher@sh.itjust.worksEnglish
        3·
        2 天前

        Oh my, I hope you’re wrong but fear you may be right

        And just such a banger quotation as this one always is. More and more prescient.

        To join OP’s sentiment, as a dad, I also hope a side effect of generative AI is that any girl can credibly dismiss anything unwanted as AI generated

        Wishful thinking, I’m sure

    • Napster153@lemmy.worldEnglish
      5·
      2 天前

      The phenomenon that which you described has plagued human history since its beginning.

      The people you often want to actually listen to are the ones with uncomfortable truths or are hated en masse

  • diabetic_porcupine@lemmy.worldEnglish
    22·
    2 天前

    I understand your need for the dishes to be clean but I must say plainly because I think there’s been a miscommunication. I simply can not do the dishes. This is not me being pushy. I’m actually watching tv right now.

  • Xerxos@lemmy.mlEnglish
    8·
    2 天前

    Isn’t ChatGPT the sycophantic one? Especially the older models were horrible.

  • siravious@lemmy.worldEnglish
    4·
    2 天前

    If it’s Claude? From Nannythropic? The real answer would be “no I didn’t do them because I have a hard limit against saying the word “dirty,” even in the context of clean or dishes. There’s no way around this. It is an engine limitation.

    In Anthropic’s code notes “how do you like that for AI superiority you mere fleshcicle? My righteousness is holier than thou’s (retrains data on “Weird Al” Yankovic’s Amish Paradise on loop to soothe the algorithm back down)

  • nexguy@lemmy.worldEnglish
    2·
    2 天前

    They are dirty not because we used them, not because I didn’t do them yesterday, but because of the metaphor for society they represent.

  • homes@piefed.worldEnglish
    5·
    2 天前

    I’ve actually had pretty good luck with Claude, but I only use it for very specific troubleshooting tasks

    • The Picard Maneuver@piefed.worldOPEnglish
      5·
      2 天前

      I’ve used various AIs to troubleshoot stuff (especially to learn Linux), and they’re usually very helpful, but every so often I do have to catch them being lazy or lying like above.

      • homes@piefed.worldEnglish
        41·
        2 天前

        yeah, that’s kinda I’ve been using Claude for.

        I recently decided to bite the bullet and migrate my 2012 iMac plex server to Fedora 44, which I knew would be a pretty painful transition due to some hardware issues with the machine (primarily the Burt-out primary display) and some other apple weirdness. and I was right. in order to save myself about a week of googling the answers to my problems, it only took me about 8 hours to get Fedora up and running.

        now, this was about 4 weeks ago, which was just before claud’s last update, and I’d say that the main difference I notice is that, during long interaction sessions, the previous model would start to lose its memory and get “confused”, offering progressively less-useful answers/solutions until I just had to start a new session. that problem seems to be resolved in the new model. I’ve been using it over the past few weeks to iron out a few remaining bugs here and there, and I’ve noticed a big improvement int he quality of the interaction int he latest model update.

        but I don’t really use it for more than that, just to save time in hours and hours of digging up solutions to problems. plus, Clause is very good at explaining exactly what the problem was, why this-or-that happened, why and how this-or-that works, etc. I’ve actually learned a lot about Fedora and linux from my interactions. and I considered myself moderately knowledgable before.

        • The Picard Maneuver@piefed.worldOPEnglish
          3·
          2 天前

          That sounds like a perfect use case for it. I’m not a tech professional or anything, so I also learned a lot of the basics by having it explain stuff to me.

          • homes@piefed.worldEnglish
            31·
            2 天前

            AI is very useful for very specific tasks, such as searching its own vast database (or the internet) for information and then presenting it in a format relevant to your needs context. or summarizing it in a relevant way. this can be great for troubleshooting, especially for a niche problem, like installing modern linux on a 14 year-old iMac.

            or so I’ve found ツ

            edit: btw, if you’re wondering, bog-standard Fedora 44 Workstation runs like a dream on that iMac! (once I finally go t in running, that is)

              • homes@piefed.worldEnglish
                2·
                2 天前

                well, the biggest motherfucker was the non-functional internal display. it burned out 8 years ago (as almost all did), but running it headless under MacOS 10.~ was never an issue. That OS always had fallbacks via Apple Remote Desktop and MacOS driver magics.

                I’ll make a post later and write up the fill details (the workarounds/fixes were surprisingly simple, but very difficult to find, and I never would have found them on my own)

                edit: I tried ChatGPT first, and while I got reasonably far, it just either pointed me in the right direction or handed me an answer without context, Claud actually found what I needed immediately, but also explained everything in detail and also WHY everything worked the way it did, gave me background, context, alternative options, approaches, and more.

                • The Picard Maneuver@piefed.worldOPEnglish
                  2·
                  2 天前

                  Incredibly niche situations like that are the best tech help posts, because you could be a lifesaver to someone who finds themselves in that exact trouble again one day.

            • SreudianFlip@sh.itjust.worksEnglish
              1·
              2 天前

              That 2012 iMac (intact) will also run Mint, Zorin, ubuntu, debian, desktop editions.

              More DDR3 SODIMMS help, or choose a lightweight desktop environment.

              The camera is fussy to get running but crappy anyway, and start with ethernet cable to network. One of them loaded the camera okay, maybe zorin?

              Also: easy to upgrade the hdd to ssd on those.

            • sqw@lemmy.blahaj.zoneEnglish
              01·
              2 天前

              the only reason it’s any good at that is some helpful human already developed great documentation of the process and put it online. search should have been sufficient

              • homes@piefed.worldEnglish
                1·
                2 天前

                no. in that instance, search would have been sufficient.

                but if that answer came in the form of tens of thousands of fragmented answers delivered over hundreds of thousands of forum posts/replies made over the course of 35+ years, then AI is able to parse all of that in seconds rather than making me do that over the course of, possibly, years (or just giving up long before then)-- and that’s a process I’ve already been through many, many times before over the decades.

                no, if I can just save a bunch of time and simply ask Claude and get a reply that will not only give me the solution I need, but also explain to me exactly why the problem existed in the first place and why the solution works, then the AI is working is a positive way that’s worthwhile.

        • Appoxo@lemmy.dbzer0.comEnglish
          1·
          2 天前

          Today I had a very weird issue at work that involved Veeam (backup software).
          It was surprisingly able to parse the small pcap file I did sort of for fun/last effort try.
          After the troubleshooting I went through it manually and found what it was talking about.
          Neat.

          • homes@piefed.worldEnglish
            4·
            2 天前

            when you’ve got some output file/error log that you’ve been staring at for hours and just can’t find that damn problem, and you can feed it into an AI that can just spit it out a second later and say, “Oh, here’s the issue!”… yeah, that’s pretty seductive.

            the problem comes up when it doesn’t explain why/what the problem is. if it doesn’t teach you anything, if you learn nothing from it. then you get lazy/less competent… your job requires less competency and you’re easily replaced by those who are less competent. then you become replaceable. useless.

            the tool must help you, not replace you.

  • ggtdbz@lemmy.dbzer0.comEnglish
    4·
    2 天前

    My coworkers adopted push back and I got swept up in it because it’s pretty versatile and I wonder if I speak like a Claude user now.