My experience certainly felt a lot different from yours, possibly because I tried to establish rules of engagement upfront. Rather than offer up options, as your 'player' did, mine was clearly trying to tell me how things played out. I will certainly be trying your more freeform approach in future as I was extremely impressed with how things played out there.
And, yes, there was certainly a consistency to the character. I would attribute that to your decision to keep things as simple as possible, allowing the AI to choose a medieval architype.
Thanks again for such a generous and entertaining contribution to the episode.
Good Stuff, Spencer. And thanks for the flattering edit of my message ( taking out all my errors and double takes). A very curious take from Joe there (one I hadn't considered, and one I think I need to ponder more on) and a similar experience to your own experience from Menion. A thing I noticed about ChatGPT; is it doesn't ask questions to establish the precise facts of a situation. Instead it seems to preempt what you mean and present multiple possible options or actions it would like to take based on what it has presumed could be the situation. This is why I say it is a flow chart combined with a data dump. As a GM I felt that it was therefore my responsibility to pick one of the branching options of the flow chart that it comes up with and lead CHATGPT down that branch or option. And I usually took the chat down the most immediate and pressing option, until I got to the 'use the stuff in the room bit'. I was thinking and acting a bit like a fighting fantasy book and I kinda already knew that the salt was the right option in the situation, but I took pity on Chat GPT and rather than just take it's first choice and penalise it for picking incorrectly, I decided to use all the options it had presented and create a dramatic narrative out of it. Against a human player I might have asked it to choose only one action and then used that to create a sense of increasing jeopardy and desperation with each failed choice, but seeing as ChatGPT had already stated what it was going to do with each object, then me trying to create a sense of increasing urgency seemed redundant. It seemed to remember it's own character as an upstanding Knight and all it options seemed to maintain and be in keeping with that personality, which I found curious and the most impressive part of the AI. But yeah. I think best used as a punch bag work out for descriptive responses ( it's use of adjectives and adverbs to create dramatic prose is impressive, even if, as you say, a bit verbose - I think the word I thought of was 'florid'). I can't imagine trying to use it to create logical or mathematical based rules systems, but I wonder if there is a 'hypertext' way to get it to use equations to, say, resolve 'to hit AC' mechanics? That is all too much like programming for me though. I think there might be more 'legs' in this topic, but it probably needs more digging into ChatGPT than I can be bothered with. My FKR tendancies won't let me look too deeply into the abyss! Take care
Thanks, Safer. Glad you enjoyed it.
My experience certainly felt a lot different from yours, possibly because I tried to establish rules of engagement upfront. Rather than offer up options, as your 'player' did, mine was clearly trying to tell me how things played out. I will certainly be trying your more freeform approach in future as I was extremely impressed with how things played out there.
And, yes, there was certainly a consistency to the character. I would attribute that to your decision to keep things as simple as possible, allowing the AI to choose a medieval architype.
Thanks again for such a generous and entertaining contribution to the episode.
Good Stuff, Spencer. And thanks for the flattering edit of my message ( taking out all my errors and double takes). A very curious take from Joe there (one I hadn't considered, and one I think I need to ponder more on) and a similar experience to your own experience from Menion. A thing I noticed about ChatGPT; is it doesn't ask questions to establish the precise facts of a situation. Instead it seems to preempt what you mean and present multiple possible options or actions it would like to take based on what it has presumed could be the situation. This is why I say it is a flow chart combined with a data dump. As a GM I felt that it was therefore my responsibility to pick one of the branching options of the flow chart that it comes up with and lead CHATGPT down that branch or option. And I usually took the chat down the most immediate and pressing option, until I got to the 'use the stuff in the room bit'. I was thinking and acting a bit like a fighting fantasy book and I kinda already knew that the salt was the right option in the situation, but I took pity on Chat GPT and rather than just take it's first choice and penalise it for picking incorrectly, I decided to use all the options it had presented and create a dramatic narrative out of it. Against a human player I might have asked it to choose only one action and then used that to create a sense of increasing jeopardy and desperation with each failed choice, but seeing as ChatGPT had already stated what it was going to do with each object, then me trying to create a sense of increasing urgency seemed redundant. It seemed to remember it's own character as an upstanding Knight and all it options seemed to maintain and be in keeping with that personality, which I found curious and the most impressive part of the AI. But yeah. I think best used as a punch bag work out for descriptive responses ( it's use of adjectives and adverbs to create dramatic prose is impressive, even if, as you say, a bit verbose - I think the word I thought of was 'florid'). I can't imagine trying to use it to create logical or mathematical based rules systems, but I wonder if there is a 'hypertext' way to get it to use equations to, say, resolve 'to hit AC' mechanics? That is all too much like programming for me though. I think there might be more 'legs' in this topic, but it probably needs more digging into ChatGPT than I can be bothered with. My FKR tendancies won't let me look too deeply into the abyss! Take care