Comment on The math of infinite quarterly progress towards the nearest wall

• Natanox@discuss.tchncs.de ⁨12⁩ ⁨hours⁩ ago

  You can not “collaborate” with an LLM as it has no agency on its own, merely the simulation of it. You use it as a tool, that’s it. It also has no inherent expertise on anything (let alone a deep one), it merely generates the closest match which is why it will confidently do total nonsense so often. To use AI as tool safely and correctly you can only ever use it in domains you already know enough to understand when the LLM completely derails. Therefore I (as an example) can use and understand it for research on coding (but not letting it generate code for production, oh my god), but could never use it for finance advise as it could tell me any kind of nonsense without me being able to pick up on it.

  You’re right on some things, like that AI is marketed and used the wrong way. It’s also made the wrong way by stealing literally the collected works of mankind, so to have reasonable approach to AI / LLMs as tool you have to be really cautious, use it carefully and quite frankly skip the media (pictures, audio) generation completely. That one is morally corrupt even in small amounts, plain and simply.

  source

  • Quibblekrust@thelemmy.club ⁨11⁩ ⁨hours⁩ ago

    You knew what he meant. Don’t be pedantic. Having “a back and forth conversation with an AI, where I also made manual edits at times, to iteratively produce on an image” is too verbose, and “collaborate” sums it up nicely.

    Especially since the conversation with the AI is literally in plain English. Do you talk to your power drill in English? Or table saw? Or toaster? No, but you do talk to an AI and have a conversation.

    Just because it’s not “intelligent” or “self aware”—or other things you definitely don’t want it to be anyways—doesn’t mean you aren’t having a conversation or collaborating. Those words apply just fine.

    source

    • Natanox@discuss.tchncs.de ⁨2⁩ ⁨hours⁩ ago

      No, they fucking don’t. These differences matter a lot especially because LLMs behave so human-like (as it was trained on us) and we got way, WAY too many people already treating them like equals, work buddies, friends or even fucking falling in love with a tool (ever heard multiple people genuinely fall in love with a power drill or a toaster, or killing themselves because the table saw told them they’re the messiah?). Words matter, especially in regards to what we call “Medienkompetenz” in german (lit. “media competency”, meaning bring educated in how to safely consume and use any kind of media or digital tool). And even more especially with AI, where not just the salesmen but even the tool itself tells us it can do shit it certainly can not while trying to abuse our emotional weaknesses.

      Collaboration (merriam-webster.com) “to work jointly with others or together especially in an intellectual endeavor”

      You do not “collaborate” with a fucking toaster. If you think you do with an AI then your Medienkompetenz is lacking and you will eventually, even accidentally, put too much trust in what is essentially a game of chance and the machine will fuck you over, on which it will just say “You’re absolutely right!” while you have to witness all the consequences of your action (an AI can’t take responsibility because it’s just a bunch of complex math). The fact this tool uses natural language doesn’t change anything except your impression of it.

      Stop subtly treating LLMs like a god damn entity, you’re fooling yourself.

      source
  • voodooattack@lemmy.world ⁨11⁩ ⁨hours⁩ ago

    Most of what you speak of is caused by this simple statement from the onset:

    You use it as a tool, that’s it.

    If you use it as a tool, then expecting anything but minimal compliance would be an irrational expectation, because it will do exactly as you ask, which is the bare minimum to satisfy your requirements.

    I got tired of that, so as an experiment, I decided to go about it differently, and made a couple discoveries about how to make it work. Turns out LLMs can’t pattern match things they don’t have in their training sets.

    source