I heard ChatGPT was a decent enough solo DM experience and I decided to try it>generic human fighter whose job was to guard a princess but failed>first session starts alright, fought some water soaked zombies and found the princess>couple of more sessions in there's a global level system which dictates which reality is true and uses human anchors in case it can't stabilize a location>another couple of sessions in I am now arguing logics with ChatGPT because I got some superpower which lets me create point and relationships between them>end up this first arc building a system where synthetical ancors replace human anchors and it took me a while to cover all edge cases>this is at level 4>explicitly say I do not want to deal with this stuff anymore>ChatGPT: "You're totally right!" ;^)>next arc is about a network of reality bending doors in which couriers must transport fragments of reality from door to door to keep the reality in check>the only way I could disrupt this was by PLACING FUCKING STEPPING STONES IN AN IRREGULAR BUT NON-OBVIOUS WAY>my fucking god>write explicity multiple times in the source of the project to not have any other reality bending stuff and all that happened earlier was an exception>ChatGPT: "You're absolutely right!" ;^)>next arc is about a "grounder" of sorts which fixes all contradictions in the universe>I had to literally do a "This statement is false" to break it
>>97837598ChatGPT is bad at being a DM because it cannot improvise, it can only reiterate that which exists.
>>97837598wow, youre retarded
>>97837618Yeah I noticed, still better than having to be a forever DM at the moment.
>>97837598Could you give mode information? Did you tell it to make an acid trip of a system or was that suddenly out of the blue?
>>97837667Apparently it was because of how I was messing with it. It started from:>a basin holding water which can be interacted with a rod>some wells also do similar things>actually these things are nodes and they can become unstable and create logically breaking visible patterns, and they can kill you>actually nodes are everywhere but as you are literally going against this system you can just sit in an unstable zone and fix it>after some training you can create points and relationship between them but beware they're pretty much computer programming but more complex>actually now I'll give you what's essenstially moderator privileges over the system after you interacted with something the NPCs told you specifically not toThe only somewhat fun part was teasing the princess as for some reason she was affected by this system and me going against it made her stabilize if I was close to her, bed included
>>978375982023 called, they want their "chatbot can be a GM" shitposting threads back
Sounds like user error
>>97837598I've only really found ChatGPT useful as a sounding board for ideas. It'll elaborate on something you pitch and then you can pick that idea for a valuable nugget to iterate on or tell it "no, I don't like that idea."
>>97838354Yeah... If I get to do another campaign I am totally trying to remake that system but without the computer fuckery.
>>97837598You could use it to run a dnd module. It follows the rules pretty well since it's all written down in the pdf.
>>97837598>i talked to a probability algorithm and it couldnt do human reasoningwow...
>>97838477I just had to throw in Godel's incompleteness theorem to collapse what's pretty much a robotI'm living the good life
>>97837598Try deepseek or gemini instead and you'll eventually run into similar problems. Hilariously, the AI bubble is being popped from the inside and they're all setting harsher use limits and weaker performance.
>>97838561You're tempting me with a Gemini DM and ChatGPT + Deepseek players campaignThis'll be so fucking good
>>97837644If no one's allowing you to be a player, anon, you should probably think about why that might be. A scarcity of GMs only goes so far. Hint: how are you finding players?
>>97838668They know I was a DM a couple of times. Most of the other groups I had back when I was a player are from high school and I lost contact
>>97837598Wouldn't play with an LLM since they can't even keep track of what they're currently writing about.I just wanted a DPS comparison between two ranger builds, and it forgot then made up things it was currently writting. In the process of writting it.Sucks at replacing people, but it ain't all bad as a tool.
>>97838354This and I've found it useful for helping me make better riddles/puzzles. I input my original vague idea/puzzle/riddle it outputs pure garbage but sometimes it generates something useful.
>>97837644No, it isn't. Not even a little bit.
>>97837598There are better ways to go about using AI this way. You went head on with zero research using the lowest common denominator tools to achieve it and you're somehow surprised you had terrible results. Instruct LLMs are terrible at creative endeavors, and AI doesn't have unlimited context memory, so you can't just keep going forever expecting full cognizance of the entirety of your campaign. You're going to have to work with back-end APIs at the very least if you want to have a remote chance at being successful at this.
>>97837618>ChatGPT is bad at being a DM because it cannot improvise, it can only reiterate that which exists.TTRPGs are reiterating the cliches of a given genre on the fly. There are plenty of reasons as to why current LLMs can't DM as well as human can, but this is not one of them.
>>97837598Another Solo ((((((Player)))))))) takes an LJUST READ A NOVEL LMAOOOO
>>97839055how is a novel like a game?
>>97839120> Panicking in Virtual Novel
>>97837598AI can never solo DM as of now, it can work as a co-DM at best, but it takes a lot of tinkering and you'll always have to curate to some degree because there will always be retarded generations no matter how good your prompt and model are. You'll probably want to use those AI roleplay chatbot tools because trying to use an AI assistant for this is pretty much akin to bashing your head against the wall, it's just not designed for that.
>>97839328I think the biggest problem is the size of the context window.Google Gemini struggles to digest the entirety of singular epic poems, like GK Chesterton's "Ballad of The White Horse" and hallucinates extra stanzas.It lacks the ability to retain and refer to that amount of information, or really contextually understand which bits are important in the moment, and which to handwave.
>>97839428Which is why you need to use tools for automatic summarization and conditional referencing (i.e. lorebooks) so that the context window can remain within the limitations of your current model. It's not something those AI assistants are designed for but roleplay LLM front-ends have been working with for years now.The problem with AI assistants is that their actual prompt includes your entire past conversation in that session (or whatever they can fit), so it's just needless bloat to your context memory compared to going directly to the API with your own tailored rules of what the prompt should look like.
>>97837598>I heard ChatGPT was a decent enoughI heard that you're a sucker.
Amateurs. You have to tell AI to not be sycophantic.Use the following prompts:Do not be sycophantic. Challenge my assumptions, point out errors, and prioritize accuracy over agreement. Be generally supportive and constructive, but do not agree with me automatically. If my ideas have merit, acknowledge that briefly. If there are weaknesses, gently point them out and suggest improvements. Avoid excessive praise or flattery. Maintain a collaborative tone while still ensuring accuracy.”
>>97839601i just use my wife. easier to correct
>>97837598>I heard ChatGPT was a decent enough solo DM experienceYou heard wrong.
