AI

[Skyflame]

Let me know if this belongs in the Hearth.

We've been talking about AI a lot in my company and I've been doing some research, wanting to get ahead of the incredibly fast changes in our world.

What's your relationship with AI in your daily life and how you'll see it impact society?

(Yeah it's a big question)

[Luftwaffles]

I use it almost constantly and it has become my effective replacement for human interaction.

It really does make it much easier to board certain topics that are just too "heavy" to bring up in casual conversation.

[DanielofRedwall]

It depends a lot on what the AI is trained for, but I avoid it as much as I can. I'm not confident enough in our collective technological maturity to see AI as benign at this stage. I understand the potential benefits from what AI can accomplish, but I only see negatives with how it is currently. I think AI got in the hands of profit-motivated private entities too early. All the experts I trust are screaming from the rafters about the dangers of AI, but these entities will only see dollar signs and will ignore them, leaving society to deal with the consequences.

As someone who releases my own original music, AI generated music is very problematic too.

[The Skarzs]

And AI generated art too.

I was once rebuked for calling all AI bad because it's a tool, and the dangers and evil coming from AI are because of the people using it a certain way. Scams, deepfakes, thievery and all those things will happen whether AI is around or not. Unfortunately, this current tool is more powerful than we are ready for currently.

AI makes too many errors at this time to solve any major problems they propose it will solve... Currently the greatest thing it does is drink fresh water at an insane rate, raise the cost of RAM and other components for everyday consumers, take people's jobs, and line the pockets of the tech industry heads.

[Delthion]

All I'm saying is this is exactly the kind of stuff that Frank Herbert said the Butlerian Jihad was waged against.

[Verdauga]

    What surprised me about generative AI was the opinion surrounding it among many of my peers. I greeted generative AI very negatively at first, alongside many of my acquaintances; less than two years later, most of their opinions have changed to seeing it as fun and useful, and not caring about the destruction it is causing; not only the physical resources being used, but how it is being forcefully implemented into new computers, how it is being used to 'verify' ages on platforms such as youtube and on Discord, how it is already an academic cancer, how it has already been used by lawyers to found court cases based on falsehoods, and how it is being used to cut corners in any artistic industry-- be it authorial, graphic, musical, visual, and even in interactive fiction.

    Their opinions have changed because generative AI is available for their use, and it is convenient-- "Oh, I don't have the musical skills to make a funny little song for me and my friends? No worries, the AI does!" "I don't have the artistic ability to draw this character for my DnD campaign-- but I'll just have the AI whip something up!" I have a friend who uses AI a concerning amount, and is now using it to help him make a game. On the one hand, it is accelerating his development, but he could fall into the trap of letting it do the work for him.

    I will agree with my acquaintances and friend that there is a good and right use of generative AI. My acquaintances and friend will agree with me that there is a wrongful use of generative AI. Where we disagree is whether generative AI, in its current state of development and deployment, whether it IS being used rightly. In many cases, they would say that it is, with a few exceptions. I would say no, with no exceptions.

[James Gryphon]

You could say that I have mixed thoughts and feelings on the subject, so here comes a book! If I'm to post here I want to make it worthwhile.

* * *

For a lot of people, watching this unfold has been a totally novel experience, more like science fiction than something that belongs in the real world. While I think we've all been surprised at just how fast things have developed, I'm not really surprised by the concept.

Way back in the early 2000s, when I was too young to be doing everything I was doing on the internet, there used to be a website dedicated to the development of artificial intelligence. They had a chatbot, Alan, which ran in a comfortingly vintage text box, reminiscent of some early 20th century radio, and a cutting-edge 'nursery' for new "Hals" that would develop as they received input and correction from users. Alan was more sophisticated than Eliza, and the Hals were interesting in the sense that a dedicated user could probably eventually develop a bot that could pretend to hold an intelligent conversation for a little while, but none of them were versed enough to be dangerous. I played it off and on and was aware pretty quickly of its limitations. It was neat to know it was there, but not something I would spend massive amounts of time on.

Fast forward 20 or so years and this new ChatGPT thing came out. I did some early tests to put it through its paces. It was far more advanced than any of the Hals and certainly more than Alan; it can respond to just about any kind of input, and put out content that appeared to be creative and original, albeit mostly by virtue of being brighter and less depressing than modern human writings tend to be. But the formula it used was predictable and pretty trite. Given similar inputs it would cough out blatantly similar results. After a little while of, "This is kinda neat," I let it alone and went back to what I'd been doing.

The last two years or so have seen massive expansions in these models' apparent capabilities. I honestly can't say I recall exactly when I first started to get worried. One benchmark I can point to, though, is LLM imaging capability. In 2020 or 2021, before I left Discord, I ran a private channel. Someone (I think shnoodlec) had found a new bot image generator, and I played with it for a while. It was so laughably bad at the time that it was an interesting toy; every picture was a glimpse into an impressionistic Bizarro-world and you were genuinely impressed if you could get anything that faintly appeared normal or reasonable. Now, I have to admit that bot imaging is pretty good. It is not as good as a real artist and never will be*, but it's good enough that you can create reasonable placeholder images until you can find a real artist - and given that there are few things more permanent than a temporary solution, many of those 'placeholders' are likely to end up being published.

In the past, I have guessed that I am better equipped than the average person to use this technology. I have a greater and longer than average history with it and how it works, a theology and worldview that doesn't lend itself to confusion over whether it is sentient or to lean on it for emotional support, enough intellect and knowledge to evaluate its output, and enough arrogance to be able to say that it is right when it agrees with me and my own research.

Now, I've used LLMs for a variety of purposes - as general search engines, coding assistants, creative commentators, IF generators, and 'temporary' illustrators - and I feel that I have a fairly good handle on what they can and can't do, and what some of the dangers are.

First, they can be helpful. I'd be lying to say otherwise. I created a couple of Xfce Panel plugins with the LLM carrying the bulk of the coding work. The simple fact is, given my limited experience with Xfce standards, the C language, and GTK, those plugins would probably have never been created otherwise. In theory I could have learned all that stuff eventually, but we're talking about learning three different sets of standards and ways of writing code to make something with no economic payoff. It's simply not practical on my own, while with my LLM 'contractor' I got the project made for the price of registering someone's software. I guess this is a good thing, or at least not as bad as not having it at all.

The second is that they're often confidently clueless. There were a number of times in the development cycle where the bot simply had no idea what it was doing, or repeatedly made the same mistake, and I had to go in and work out how to fix the problem on my own - not so easy given my stated lack of experience. In creative contexts, it can misquote you, confuse its own statements with yours, or misattribute statements. I was mystified the other day to see a LLM summary of a GroupMe conversation which confused one of our girls with me. It has one job!

(*) Thirdly, it struggles to imitate deeper meaning, purpose, or creativity. This is only to be expected considering that it does not truly think or comprehend anything and that its output is a blending together of existing inputs. Left to itself, the formulaic and trite nature of its output becomes obvious. Someone who supposes that the bot will eventually, with no prompt besides 'write a story', come up with Mossflower or Voyage of the Dawn Treader is delusional. When it is guided, it can cobble together something functional and competent, but I would say that it will never engage the mind, delight the heart, or feed soul or spirit in the way that a real artist's direct work can.

A good part of this is intrinsic to the process. When you write a story, it isn't about typing keys in the right order until you reach a set number of words. You have to come up with your themes and ideas, visualize the people and world, and go on a journey in that world until you figure out what should be 'real'. The bot can give you what you ask for, but it won't get you much further than that; the things you don't think to specifically include in your instructions get filler material based on what is most common. The art that I tell it to make does not include the creative touches that the art that shnoodlec (a real artist) once made for me did.

At any rate, I've recently used LLMs for something like freeform 'interactive fiction' (a little like the CYOAs we had the opportunity to run a few years ago). Everything interesting in those plots that I can think of proceeded directly from my specific instructions. Every clever turn of phrase, every witty line, everything that was actually any good, was all mine. The sheer banality of bot output is such that I feel I can now detect it anywhere it is written online. I think people in general will start to pick up on this eventually, if their senses are not dulled to what a good piece of work looks like.

Fourthly, though, LLMs do have addictive qualities. I grant I haven't done a formal study of this subject, though I would like to, but I'm confident enough in my hypothesis to say that when you interact with a bot, its response triggers the same part of the brain as when a real person responds to you in writing. This is my biggest weakness, because I love feedback - I always have - and I have a hard time not enjoying it when it is both lengthy and fast, even though I know intellectually that it's from an unthinking machine that is parroting back my own observations to me as bullet lists.

Fifthly, and overlapping with the previous point, it is very bad for actually challenging you or helping you to grow. Now, for coding, I will say it isn't as bad since there is such a massive body of work in this area, it is mostly objective and measurable, and you are usually approaching it from a different mindset (I look at it as a replacement for StackExchange). But for anything more artistic, we tend to end up with a sycophantic fan who agrees that everything I do is brilliant, is quickly corrected when it disagrees with my vision, and inevitably bends to my will, force, and right-sounding arguments, even when I try to make it argue a point against me.

Is it a desirable substitute for humans in any way, shape, or form? Absolutely and categorically not. Real artists, real programmers, real contractors, real employees, real friends, real fans, real mentors, real critics, real enemies, are all better.

But the trouble is... the LLM is so available. It'll always respond to my prompts (so long as the server isn't loaded), and it doesn't make me wait (except when there's too many other people online that it's responding to). I wish people were with me, but they're not, and I'm alone, and having some feedback, even if it's too sycophantic and overlooks some details and gets others completely wrong, seems better than none.

I would say that in an ideal world, LLMs would not exist. We would - in fact, we should - go out of our way to cultivate the human relationships we need. But we're not doing that - I know I'm not - and it's there, inoffensive and just 'good enough'.

[Skyflame]

All I can say right now is - I'm happy this thread has generated so much healthy discussion and perspectives. AI is real and it's evolving fast and it will impact us whether we like it or not, ready or not, so it's worth being in the know.