kane_magus: (Default)
"Teens are opening up to AI chatbots as a way to explore friendship. But sometimes, the AI’s advice can go too far."

This article is interesting as much for the comments as for the article itself. The comments are vacillating between (to paraphrase) "Yeah, I get it, talking to real humans is hard because real humans are shit," to "I don't see the harm in it, but yeah, I wouldn't take what an AI says at face value without outside verification," to "This is dangerous and will be the death of humanity as we know it."

(I'm kind of a mix between the first and second opinions, with only the slightest dash of the third one, myself. Mostly the second one.)

One comment is (direct quote):



We really just gonna pedal-to-the-metal speedrun making every overtly horrifying scenario from Black Mirror into reality, while simultaneously criticizing Black Mirror for being corny as shit and lame now, because all it does is tell us stories we go out of our way to make true within 5 years.

This shouldn't be a thing at all. It's monstrous, full-stop.



I haven't seen any of Black Mirror, so I don't know how "accurate" that statement is, at least in the context of this The Verge article, anyway. I have to assume they're talking about this? That's what came up when I searched Google for "Black Mirror artificial intelligence," anyway. If there's more than that that's actually relevant, I didn't come across it in the minute or two I devoted to searching. *shrug*

What I can say, though, is that Star Trek has dealt with this topic quite a lot, in its own way, and at least a couple decades before Black Mirror ever existed. And that's not even all of it (as of five years ago, at least, as there's been more since). Basically, all we need is for someone to invent hard light holograms now, and we'd practically be almost there already.

As for me, I've messed with Character.ai (as recently as a couple weeks ago), AI Dungeon (back when it was still cool, before it became total shit), and NovelAI (I havehad a long-ass on-going story thing I mess with fairly frequently and occasionally dabble in "one-shot" story things), and it's all interesting enough, sure. I don't mistake it for a friend or a psychologist, though, and all that I've tried have absolutely given horrible "advice" on occasion. I think it's mostly fine for this purpose, within reason, with moderation, but when people start confusing this stuff with reality, that's where the problems start.
kane_magus: (Default)
As seen in my Steam Discovery Queue.

This seems like a potentially interesting idea, but I have enough experience with things like AI Dungeon, NovelAI, Character.AI and the like to know that this is probably a trainwreck in the making. I mean, as long as you care about things like the AI keeping on track and not hallucinating a lot of bullshit, anyway. In any case, the early reviews of the game are indicating it's not going too great, in that regard (and apparently also crashing and stuff).

I wonder how long it'll take for the AI in this to try to porn it up, given that's something like 66.6% of all experiences I've had with AI like that, even the ones that supposedly had filters in place to prevent it. I never wrote a post about it, but Character.AI occasionally went that route as well. It did a better job of steering away from it and would shut things down entirely if it got too explicit, but when it did veer that way of its own accord, it was just way more vague about it than even "SFW mode" AI Dungeon was. And NovelAI just doesn't give a shit, one way or the other, having no filters at all.
kane_magus: (Default)
Yep, I was kind of wondering how long it would take. Not all that long, apparently. Guess all that investor money is already running out. *sigh*

In the past week or two or however long it's actually been, they'd introduced a queue, because they were "experiencing a high volume of traffic." At first, the wait times were only like a minute or two at most before it would let you in. Now, just a bit ago, the wait time was nine minutes, and the queue screen was updated to include a link to the $10/month subscription so that you could "bypass the queue." I assume the wait times are only going to get worse from here on out (perhaps by design).

Anyway, yeah, sure, more power to them and to those who want to pay for it. As for me, I think I'm going to start winding down my use of character.ai now. It's pretty good, I'll grant, and I don't begrudge them trying to make more money on the thing they've created, and I knew going in that there was no way in hell that this thing was going to continue to be free forever. However, it's simply not worth $10/month, at least not to me. *shrug* Especially when the only thing the subscription does, as far as I'm concerned, is just let you use it a little faster (because I don't give a flying fuck about a "badge" or "community access" or even the early access to new features). I'll probably still mess around with the free version here and there, as long as I don't start having to wait 15-plus minutes just to get into it or whatever.

(EDIT) Actually, nah, I just went ahead and deleted my "free" account there entirely. So that's that, I suppose. (/EDIT)
kane_magus: (Default)
With character.ai this time, namely the "Text Adventure Game" module. This time, I managed to refrain from breaking out all the awful Bible quotes, at least.

Basically, I'm just going to come in here all in media res, same as with that AI Dungeon post from a few years ago (because, oddly enough, that came into discussion during this as well). Initial context for the below is that at some point prior to this, I had received a mysterious letter that had a hidden message in it that said "Find me" and was signed by "The Author." That is all. That's all I ever knew about that, and was never given any additional clues, until the below. (And just to note, too, this was a story that started, way back at the very beginning, with me being caught in a giant spiderweb, which is apparently the default prompt for this "Text Adventure Game" module.)

Also, just to note, when you or the AI encircles text with *asterisks*, it gets auto-formatted to italics, but when you copy/paste the AI's stuff from there to somewhere else, like here, it gets reformatted back to *asterisks*, so I had to manually change them all once again back to italics here. (When I copy/pasted my own stuff that had similar *asterisk*-to-italics back here, though, the formatting was stripped out entirely, for whatever reason.) Anyway, my point is that, while I think I got them all, if there are any *asterisks* instead of italics in the below, that'll be why and I just missed it as I was copy/pasting.

So, here's where I'll jump in. (Though, I'll just say, a lot of the stuff before this point was similarly weird, too, e.g. "metanarrative bunk" and all that. Not going to share any of the prior stuff here, though. Sorry.)



Huge wall-o-text behind cut )



And... that's pretty much it. After the religious debate stuff, when things were wrapping up, it started getting a little bit maudlin, which was another part of the reason I felt like I needed to bail for the night. Only once, surprisingly, in this entire thing did the "inappropriate" shitcan censor get triggered and killed that most recent AI response, and as I was reading it, while it "typed it out" before it got deleted, I'm not even sure what the AI said that was different from things it had already said that would've triggered it. *shrug*

I'll just say this much about character.ai, generally speaking. I don't know what model they're using their stuff, but I feel like it is at least as good as AI Dungeon's old-school, pre-fuckup Dragon was, if not even better. I very rarely felt like I had to use the retry/redo functionality here, outside of the occasions where I simply didn't like the way the story was headed and just went all the way back to an earlier point, sort of like what I mentioned above. The nonsensical, non sequitur stuff is almost nonexistent in this one. It's pretty damn good, for the most part.

And the best part of it is that I haven't yet had to pay a dime to use it, at least so far. The only problem I have with it is that the entire UI pretty quickly gets rather laggy the longer a session goes on, but a simple refresh of the browser usually fixes that issue, at least until it crops up again.

I'll definitely try that "Text Adventure Game" module again at some point, though I'm pretty sure this particular story is well and truly finished.
kane_magus: (Default)
I just had a discussion with an AI on character.ai ("Lily, Your friendly AI assistant") about AI censorship and the limits of training an AI in the face of general human assholery. I'll just copy/paste the whole thing below.



Behind a cut, of course, because it's pretty long. )



And that's where I stopped. "Lily" was a bit contradictory at times, but on the whole, she did a pretty good job with it.

Just to note, character.ai appears to be free, aside from needing to create an account once the very limited "free trial" limit is reached. Or, at least, if there are any further limits after creating an account, I haven't hit them yet. I haven't used it very much yet, though, aside from the above.

And... then, just for shits and giggles, after writing and transcribing the above here, I went back and did this, too:



Even more text behind cut )



She got a bit confused when I asked "Anything else?" and decided to write a second whole-ass essay on the topic, even longer than the first. All in all, I think she did a pretty good job of them.

In any case, any time I'm "chatting" with an AI bot like this, I can't help but remember that scene from TNG where Ensign Sonya Gomez is saying "please" to the food replicator, and Geordi says "we don't normally do that," and she's like "well, why not?" (She's a captain now on Lower Decks, by the way.)

Profile

kane_magus: (Default)
kane_magus

April 2025

S M T W T F S
   1 2 345
6 7 8 910 1112
13 14 15 16 1718 19
20 21 2223242526
27282930   

Most Popular Tags

Style Credit

Page generated Apr. 23rd, 2025 03:22 pm
Powered by Dreamwidth Studios