kattekrab
Hey... help me out?
How are you using (or avoiding) AI?
Some context: I'm curious. I'm seeing some people embrace and experiment with it, I'm seeing some raise red flags, I'm also seeing eye-rolling at the hype.
I'm also seeing innovation, and pushing what's possible, and documented real harms done.
All of which is feeding my curiosity, and caution.
So what do you see? Are you all in? Avoiding? Other?
djsumdog
The vast majority of people don't understand the basics of how LLMs work, or that they're just next word (next token) prediction machines. Once you get a grasp of that, it makes a lot more sense how they [don't] work. Whenever an LLM gets something "right," it's by accident. For simple things, that accident rate might be like 80%, which is good enough for people to be impressed by a chatbot.
But when you see interviews where they ask a chatbot how much it has grown since the start of the interview and it says "50x" or whatever, it's literally bullshit. The model is static. It doesn't change. Now OpenAI and others might do some nightly updates or whatever, but the model isn't changed in the prompt phase. Why does it say that? Who the fuck knows. It ingested a shit ton of Sci-Fi books during the training phase?
3Blue1Brow does a really good brief video on LLMs: https://youtu.be/LPZh9BOjkQs
and he also did some deep dives if you know computer science and remember the back propagation algorithm from machine learning class:
...the real danger of LLM is not the "AI" part, but the fact that people are trusting the bullshit machines. The machine is not "hallucinating." It's just predicting the next part of the word using a transformer with a series of attention and perceptron blocks. It can't set goals. It can't reason. It's just predicting what text looks good to a human based on a few hundred people hired to sit at desks for a year clicking on the auto generated text that looks least retarded, for hours a day.
tyil
@kattekrab@aus.social So what do you seeI see a corporate funded hype for a technology that isn't real, and likely won't be real for a long time. LLMs are not "AI", as they possess no intelligence whatsoever. Corporations push it in whichever way will be most appealing, and the majority of the human population really likes the idea of AI, and does not have the technical skills to understand it is not AI. They've spent a lot of money on it, and they want to make a return, which is why its pushed so aggressively.
From my closer circles, who are technically competent, I see much more push-back. Part of that is due to the horrible experiences they get from scrapers, obviously. But those of them that did try all come to the same conclusion too. LLMs are fine to be used in the same way a new intern can be used. Some very simple tasks that you already know how to solve can be handed off to it, but you'll always have to re-read it to be sure they didn't subtly fuck it up.Are you all in? Avoiding? Other?For now I'm avoiding, since the creation of LLMs is absurdly harmful to the open Internet, and usage doesn't provide any significant benefit. Since I already avoid proprietary software where I can, avoiding LLM integrations is, so far, pretty doable. The main place where I do see LLMs being pushed where I'd rather not see it, however, is online search, and I've noticed its become noticably worse over the past few years because of it.
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.