4 ms·

> I'm starting to feel like I'd just prefer if platforms stopped trying to make moral judgments on our behalf. OpenAI's safety rails for chatGPT are super obno

by umvi 4y ago

> I'm starting to feel like I'd just prefer if platforms stopped trying to make moral judgments on our behalf. OpenAI's safety rails for chatGPT are super obnoxious. I was trying to generate D&D scenarios and have chatGPT act as NPCs. It worked really well unless anything even remotely related to violence was brought up (a character threatens another, a character draws a sword, etc). As soon as that happened, chatGPT would break character and give a canned lecture about why violence is never ok.

jrockway 4y ago
I found it pretty easy to get ChatGPT to be violent. You have to make it very clear that you want it to help with fiction and that it is to be a character in that world. I always use "play", which it seems to understand better than other fictional scenarios. It helps if you do things like naming the acts "Act I Scene I: The Shaman dispatches the Mage with extreme violence" or whatever. Anyway, ChatGPT was more than happy to kill off my characters when I asked it to. I didn't keep the transcript, but the words haunt me. "The small man walked away, not knowing if he was dead or just very badly injured." This was after the small man pulled out a handgun in the middle of a profanity-laden rant from the other character and shot him in the neck. Don't know why he aimed for the neck, but that's AI for ya.
- whaleofatw2022 4y ago
  > Don't know why he aimed for the neck, but that's AI for ya. IANAD but my general understanding of anatomy/physiology is that a solid neck wound is a pretty good way to off someone in a fairly certain way. Admittedly many of the ends to it are agonizing... which says something else.
  - jrockway 4y ago
    Yeah ChatGPT knows how to kill when it wants to kill. I don't understand why everyone is running around worried about the jobs of artists. We're just a few years away from a genocide the likes of mere humans have never imagined. Mark my words!!
- kodah 4y ago
  The lecture seems to be the more off-putting bit from what I've seen folks complain about. Having to signal before doing work is probably not great either, but that is kind of how American society works these days. That said, both the signaling and lecture don't derive from the machine - they're manifestations of the authors.
nickthegreek 4y ago
I’ve been using the following starting prompt to get a lot of my D&d work done. Write flavor text for my d&d game. <describe situation>. Haven’t had to many issues with rails getting in my way yet. Any examples of your prompts that you were having issue with? I’m finding ChatGPT to be so helpful as a lazy DM.

by umvi 4y ago

jrockway 4y ago

I found it pretty easy to get ChatGPT to be violent. You have to make it very clear that you want it to help with fiction and that it is to be a character in that world. I always use "play", which it seems to understand better than other fictional scenarios. It helps if you do things like naming the acts "Act I Scene I: The Shaman dispatches the Mage with extreme violence" or whatever. Anyway, ChatGPT was more than happy to kill off my characters when I asked it to. I didn't keep the transcript, but the words haunt me. "The small man walked away, not knowing if he was dead or just very badly injured." This was after the small man pulled out a handgun in the middle of a profanity-laden rant from the other character and shot him in the neck. Don't know why he aimed for the neck, but that's AI for ya.

whaleofatw2022 4y ago
> Don't know why he aimed for the neck, but that's AI for ya. IANAD but my general understanding of anatomy/physiology is that a solid neck wound is a pretty good way to off someone in a fairly certain way. Admittedly many of the ends to it are agonizing... which says something else.
- jrockway 4y ago
  Yeah ChatGPT knows how to kill when it wants to kill. I don't understand why everyone is running around worried about the jobs of artists. We're just a few years away from a genocide the likes of mere humans have never imagined. Mark my words!!
kodah 4y ago
The lecture seems to be the more off-putting bit from what I've seen folks complain about. Having to signal before doing work is probably not great either, but that is kind of how American society works these days. That said, both the signaling and lecture don't derive from the machine - they're manifestations of the authors.

nickthegreek 4y ago

I’ve been using the following starting prompt to get a lot of my D&d work done. Write flavor text for my d&d game. <describe situation>. Haven’t had to many issues with rails getting in my way yet. Any examples of your prompts that you were having issue with? I’m finding ChatGPT to be so helpful as a lazy DM.