I kind of hate AI generally, but I do use it for one thing: when I code (by hand!), I use it as a “turbo manpage” - meaning it’s documentation that understands fairly well what I’m looking for instead of being passive documents I have to read.

But I’ve come to realize that it doesn’t make life easier so much because it comes up with the technical tidbits I need quicker, but because I don’t have to deal with the hassle of visiting a dozen technical websites and dodge their numerous trackers, ads and other pointless Javascript eye-candy and other nonsense. Not to mention, I don’t have to download megabytes of crud just to read a two-line technical post - not that bandwidth is expensive nowadays, mind you…

So for instance, let’s say I have this weird problem with a C library: pre-AI, I’d search the error message (preferably not with Google, but sometimes with Google if the problem was obscure enough), and it would give me a bunch of links to Reddit, Superuser, GitHub… And then I’d visit each of those sites, unblocking their stupid abusive scripts one by one in uBlock Origin until they rendered something readable with the information I wanted to read in it somewhere. Over and over and over again, until I found the problem.

With AI, I ask it what the error message is, and it goes out and browses the sites for me. It’s not my browser going there, it’s duck.ai’s crawler, and I don’t have to deal with the abuse from the pages I used to visit myself because AI does it for me.

Those sites complain loudly that AI is killing their traffic and their advertising revenue. But you know what? After reflecting on their plight, well… cry me a fucking river. Visiting most of those sites is like chronic pain: you live with it because you have to, but the moment it stops, you realize just how painful it is when it starts again.

I do sympathize with the few sites that aren’t (too) abusive. But all the others can fuck right off: they don’t get to track me - at least provided I’m careful with what I ask AI - and they don’t get to ram their ads down my throat.

Of course it’s far from ideal: now I’m tracked by a single actor - the AI agent and its operator - and it’s really unnerving. But at least it’s more comfortable than browsing the regular web. But I do try to ask it only generic questions with non-identifiable information in my questions.

  • TranquilTurbulence@lemmy.zipEnglish
    4·
    14 hours ago

    I would love to have more of these kinds of conversations, but judging by the downvotes, most Lemmy users disagree.

    Anyway, LLMs have changed many things, but source criticism still hasn’t gone anywhere. People use them as if they’re truth engines, ignoring how often they hallucinate. If you’re not qualified to verify the factuality of the answer, the LLM will fool you disturbingly often.

    Programming and other technical matters are a bit different though. You don’t have to be an expert of any kind to verify the answer. If the command doesn’t run, the library doesn’t exist or the compiler spits out errors, you’ve just got plenty of verification. The LLM hallucinated some nonsense, and was caught before any real damage could happen.

    In other matters, verifying the answer would take time, effort and expertise. Avoiding all of that is kinda the whole point why people outsource their thinking to an LLM. People make these kinds of decisions, because it’s just so convenient. As you’re avoiding modern internet cancer by letting the crawlers take the hit on your behalf, you’re doing so out of convenience. Can’t really blame you though. Large parts of the internet is a bout to reach terminal stages.

    As you’re using Duck.ai the privacy side of it is probably as good as its going to get. However, it still has broader ripple effects. Duck.ai is using GPT, which means that OpenAI is getting a signal to keep on developing GPT, violating your privacy elsewhere, and spewing out CO2. The use of LLMs isn’t limited to just privacy matters, since it involves several ethical questions as well.

    Side note: AI is a very broad topic, and it covers good and bad stuff. LLMs are a far more specific topic, and it involves some unique downsides absent in most AI applications.