Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

LLM based agents as Dungeon Masters

studenttheses.uu.nl

Post reply on HN
Print view View original

111–120 of 144 posts

Re: LLM based agents as Dungeon Masters

#111

Post by sentimentscan »

empath75 wrote: post #13

The main problem with doing this on chatgpt is that it is _relentlessly_ cheerful and it's almost impossible for it to produce any kind of interpersonal drama or threat or danger. I tried to get it to run space opera about a diplomatic mission to a newly recontacted planet, full of Game of Thrones-style intrigue and plotting and scheming, and it was like playing with a cheerily optimistic new UN intern that completel…

Yep, in my game it tried always to "save" player it was impossible for LLM to jail or kill a player to end a game. I had to explicitly end it by game logic, not by LLM.

Re: LLM based agents as Dungeon Masters

#112

Post by thom »

PaulHoule wrote: post #86
thom wrote: post #81

Earlier quoted context omitted.

I DM for early teens and I would kill for a D&D party that took it all too seriously. :)

Try a simpler game. I think a lot of people think TTRPG = D&D and the world is worse off for it. There are numerous tactics that work to face the "not serious" problems you have in a game for "young adults". For instance, serious players get resentful when somebody not serious comes in late, forgets their character sheet, etc. You could lose either or both of them, but with a little prep you don't have to. In Paranoi…

I think whatever rules we tried to play, the nob jokes would persist.

Re: LLM based agents as Dungeon Masters

#113

Post by BerislavLopac »

thom wrote: post #81
PaulHoule wrote: post #43

Earlier quoted context omitted.

Personally I can't stand to run a D&D game as players take it all too seriously. I will run Toon , Call of Cthulhu or Paranoia any day, the latter without a scenario as I can kill off their all their clones in my very dangerous Alpha Complex and have them rolling on the floor laughing the whole time with a help of a stack of prerolled character sheets and some random encounter tables. (I'd expect an LLM to be able to…

I DM for early teens and I would kill for a D&D party that took it all too seriously. :)

Try Tales from the Loop -- https://freeleaguepublishing.com/games/tales-from-the-loop-r...

Re: LLM based agents as Dungeon Masters

#114

Post by devmor »

bongodongobob wrote: post #95
devmor wrote: post #69

Like others, I don't really understand why anyone would desire this. D&D is a social activity, why would I want to remove the most core human component of a social activity. It'd be like someone who enjoys going to a sports bar with friends deciding they'd rather sit next to a vending machine.

Do you take issue with Baldur's Gate 3? Video games in general? Why not?

Baldur's Gate 3 (and other games of its nature) are narrative experiences crafted by writers to convey a story. I don't play BG3 when I want to play D&D, and when I do want to play something like BG3 I certainly wouldn't be interested in a machine generated amalgamation of other people's ideas with no thought or emotion behind it - that would be fundamentally pointless and a waste of my time - like reading Lorem Ipsum instead of a novel.

Re: LLM based agents as Dungeon Masters

#115

Post by MiddleEndian »

empath75 wrote: post #27
erehweb wrote: post #18

Earlier quoted context omitted.

Reminiscent of the Twilight Zone episode where a criminal ends up in the afterlife. Spoilers in link https://en.wikipedia.org/wiki/A_Nice_Place_to_Visit

The old gpt2 engine that used to drive AI Dungeon had the opposite problem, it was obviously trained on a ton of really bad fan fiction and would frequently either shift into filthy erotica or some kind of hyper violent serial killer story, no matter how innocently the story started.

I loved AI Dungeon for generating funny/stupid fanfiction style adventure/action/fantasy/scifi stories. It could keep track of characters and settings, have them return with motivations. And then something unexpected would happen. Provided me with a lot of entertainment.

But my problem with using it as a DM is that you have to police yourself. I could just declare that I did something and it would automatically succeed. There were no rules or antagonism in the system. It was more of a collaborative storytelling tool than a game.

To be fair I haven't used it in years so I dunno how it works nowadays.

Re: LLM based agents as Dungeon Masters

#116

Post by jhbadger »

2099miles wrote: post #3

This is so interesting that people want this. This is beyond the guy saying “people don’t like making music so we made ai so it’s easier and faster to just prompt it to make music” . Like in my experience the best part of d&d is the dungeon master who knows their friends making a story and details specifically for fun and the fun of the group. And the interactions between players and the dm is where I derived the mos…

One of the growing trends these days is solitaire RPGs -- where people play alone as both player and Game/Dungeon Master. Typically this isn't done by AI but by rolling on various tables (sometimes called "GM Emulators"). Not everbody has the time or even enjoys the extrovert pressure of a traditional social RPG. A related concept is the "cooperative RPG" which does have multiple players but no DM so everyone can play.

Here's a list of some of the ones available (about 10 years old but it gives you the idea)

https://rpggeek.com/geeklist/181957/list-of-solitaire-soloab...

Re: LLM based agents as Dungeon Masters

#117

Post by emptysongglass »

int_19h wrote: post #102
empath75 wrote: post #13

The main problem with doing this on chatgpt is that it is _relentlessly_ cheerful and it's almost impossible for it to produce any kind of interpersonal drama or threat or danger. I tried to get it to run space opera about a diplomatic mission to a newly recontacted planet, full of Game of Thrones-style intrigue and plotting and scheming, and it was like playing with a cheerily optimistic new UN intern that completel…

Indeed. But for stuff like this, ChatGPT is overkill. It's better to get a dedicated RP finetune of LLaMA, Qwen, or some other open weights model (you can still run it in the cloud if you don't have hardware to do so locally). There are enough finetunes around by now that you can "dial in" how dark you want it, but for some examples: https://huggingface.co/jukofyork/Dark-Miqu-70B https://huggingface.co/SicariusSicari…

Just curious, how do you keep up to date on these models? Is there a community out there that discusses them?

Re: LLM based agents as Dungeon Masters

#118

Post by Philpax »

ge96 wrote: post #23
2099miles wrote: post #3

This is so interesting that people want this. This is beyond the guy saying “people don’t like making music so we made ai so it’s easier and faster to just prompt it to make music” . Like in my experience the best part of d&d is the dungeon master who knows their friends making a story and details specifically for fun and the fun of the group. And the interactions between players and the dm is where I derived the mos…

I was watching a video on YT, heard a cool song, tried to soundhound it, turns out it was an AI generated song, I felt like I was "owned" damn.

If you enjoyed it, does it matter?

Re: LLM based agents as Dungeon Masters

#119

Post by Philpax »

CuriousCosmic wrote: post #32
Kuinox wrote: post #26

Earlier quoted context omitted.

Neuro-sama, an AI VTuber streamer, manages to do this. There are topics brought along multiple streams and manages to remember details of previous streams.

With soemthing like Neuro-sama I wouldn't doubt that Vedal is doing a lot of liftwork behind the scenes to keep sufficient context available to Neuro-sama's LLM.

And with the right kind of engineering, that could be automated too.

I don't think the results would be as good as they are now - Vedal is bringing his own perspective into it - but once you have a proof of concept, you can improve it.

Re: LLM based agents as Dungeon Masters

#120

Post by Philpax »

emptysongglass wrote: post #117
int_19h wrote: post #102

Earlier quoted context omitted.

Indeed. But for stuff like this, ChatGPT is overkill. It's better to get a dedicated RP finetune of LLaMA, Qwen, or some other open weights model (you can still run it in the cloud if you don't have hardware to do so locally). There are enough finetunes around by now that you can "dial in" how dark you want it, but for some examples: https://huggingface.co/jukofyork/Dark-Miqu-70B https://huggingface.co/SicariusSicari…

Just curious, how do you keep up to date on these models? Is there a community out there that discusses them?

https://old.reddit.com/r/LocalLLaMA/ is the place to be, but it's a subreddit with an eclectic population, so apply some caution when browsing it at work.
Post reply on HN