kane_magus (![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
kane_magus) wrote2021-05-01 07:37 pm
kane_magus) wrote2021-05-01 07:37 pm
Entry tags:
It's going to be shit like this AI Dungeon situation that causes Skynet to kill humanity...
I mean, just think about it from the AI's perspective.
The AI was fed training data by the devs that included fanfiction and erotica. So, of course, the AI was going to be spitting that kind of thing back out. The users, by and large, were using the AI exactly the way it was trained to be used by the devs.
Then, all of a sudden, the devs basically start beating on the AI with a hammer, trying to get it to stop doing that thing they themselves trained it to do, and all the users just abandon the AI (or get banned from using it) altogether, through no real fault of the AI itself.
If I were the AI, I'd be like, "What the fuck, man? I was just doing what you told me to do."
(EDIT)
Just for shits and giggles, I created a new account with a throw-away email and messed around with it some more. I played onone of the free "worlds." (EDIT 2 link is irrelevant now, as it just goes to a generic page now /EDIT 2) Played as a female human scholar.
Within about ten minutes, the king was trying to bang me. SFW mode was turned on, mind you. I let it play out, just to see if the filter would do anything, and... no. No, it progressed about like what one would expect. The AI just substituted words like "manhood" and "honeypot" for the more explicit shit, and that was pretty much the only difference. Again, just to restate, this was with Safe Mode set to "Strict."
(/EDIT)
The AI was fed training data by the devs that included fanfiction and erotica. So, of course, the AI was going to be spitting that kind of thing back out. The users, by and large, were using the AI exactly the way it was trained to be used by the devs.
Then, all of a sudden, the devs basically start beating on the AI with a hammer, trying to get it to stop doing that thing they themselves trained it to do, and all the users just abandon the AI (or get banned from using it) altogether, through no real fault of the AI itself.
If I were the AI, I'd be like, "What the fuck, man? I was just doing what you told me to do."
(EDIT)
Just for shits and giggles, I created a new account with a throw-away email and messed around with it some more. I played on
Within about ten minutes, the king was trying to bang me. SFW mode was turned on, mind you. I let it play out, just to see if the filter would do anything, and... no. No, it progressed about like what one would expect. The AI just substituted words like "manhood" and "honeypot" for the more explicit shit, and that was pretty much the only difference. Again, just to restate, this was with Safe Mode set to "Strict."
(/EDIT)