Chat GPT, it will change everything if implemented.

[Don Cool]

I guess you can´t change the main story line a lot but everything below that could be changed and quests would be handed individually depending on how ChatGPT or any other similar software chooses to go about your questions. I asked ChatGPT to pose as a weapons merchant in Faerun, even asked for quest and he/she gave me one.




Me: I am looking for a longsword, something with a bit of mmmpf

Dwarf Salesperson: Ah, a longsword with some extra "mmmpf," you say? You've got a discerning eye for weaponry, my friend! Let me show you something special.

(The dwarf carefully places the Flamestrike Blade back on the counter and reaches for another sword, this one with a hilt adorned with fiery motifs. As he draws it from its scabbard, you can feel a faint heat emanating from the blade.)

Dwarf Salesperson: Feast yer eyes on this beauty, the Inferno's Edge! Crafted by master smiths and imbued with the essence of fire, this longsword is a marvel. Each strike carries the fury of a fire giant's wrath, dealing extra fire damage to your foes. It's balanced for both precision and power, and it'll make sure your enemies remember the day they crossed your path.

(The dwarf hands you the sword, and you can feel the warmth in your hand as you grasp it.)

Dwarf Salesperson: What do you say? Ready to make this fiery masterpiece your own and bring some serious "mmmpf" to your adventures?

Me :I could mine some materials for a new sword, don´t have that much gold, wouldn´t you be able to cut off a bit of the price if I tossed in a good amount of raw materials?

Dwarf Salesperson: Aye, I understand the value of hard-earned materials, and we can certainly work something out, my frugal friend. If you've got some quality raw materials to contribute, I can offer you a discount on the price of the Inferno's Edge.

(The dwarf looks thoughtful and inspects the materials you have brought. It appears to be a fine-quality steel ingot and some rare, enchanted gems.)

Dwarf Salesperson: These materials are of good stock, no doubt. I reckon they'll serve as a solid foundation for a new sword. I'm willing to knock off a fair amount from the price of the Inferno's Edge in exchange for these. How about this: with your materials and a reasonable sum of gold, we can make a deal that'll satisfy both of us. What say you?

Me: Do you have anything else I could help you with? ... you know, a missing daughter, a cheating wife that needs spying on ... or any other task?

Dwarf Salesperson: (He leans in closer, his voice dropping to a hushed tone.) Well now, it just so happens that I do have a little matter that could use some discreet attention. My cousin, Grimli, has gone missing while on a journey to investigate a rumored vein of rare mithral deep within the Spine of the World mountains. I'm growin' worried about him, and I'd be willin' to offer you a handsome reward if you'd be willing to look into his whereabouts and ensure he's safe.

(He produces a small, weathered map and slides it over to you.)

Dwarf Salesperson: Here's a map of the region he was last seen in. If you can find any sign of Grimli or uncover what happened to him, I'd be greatly indebted to you. Just remember, discretion is key. We dwarves don't like to air our family matters in public, you understand?

(He gives you a knowing look, waiting for your response.)

[Rex is back, baby!]

No. I'm against ai in all forms.

[Aria Athena]

To be honest, this doesn't sound like AI to me. I mean, "my frugal friend"... what?

But I am also against any form of AI.

[WeirdWizardDave]

Originally posted by Rex:

No. I'm against ai in all forms.

You're really going to hate the future then.

[Rex is back, baby!]

Originally posted by WeirdWizardDave:

Originally posted by Rex:

No. I'm against ai in all forms.

You're really going to hate the future then.

I won't be alive long enough to hate it.

[Vel'Rathis]

Chat GPT is not true AI, its developer Moderated in response rather than a true LLM. Unfortunately that means it has the same political bias as its San Francisco based developers.

[Faefu]

Discord Clyde:

"that sounds like a cool idea! using ChatGPT or similar software to dynamically generate quests and interact with NPCs could definitely add a new level of immersion and customization to the gaming experience. it would be interesting to see how it could be implemented in different game worlds like Faerun."

[WeirdWizardDave]

Originally posted by Rex:

Originally posted by WeirdWizardDave:

You're really going to hate the future then.

I won't be alive long enough to hate it.

Sorry to hear that.

[Capo Rip]

You should play with it some more to see it's limitations. The current iteration of ChatGPT doesn't know what facts are and doesn't build permanent or consistent narrative. It knows what a DM and PC player interaction looks like but doesn't really know beyond the form what it is. It writes out answers it "knows" appear correct, not answers that it "knows" has the correct facts. As is. ChatGPT is bad at this stuff. Good RPG games should be consistent.

[WeirdWizardDave]

Originally posted by Capo Rip:

You should play with it some more to see it's limitations. The current iteration of ChatGPT doesn't know what facts are and doesn't build permanent or consistent narrative. It knows what a DM and PC player interaction looks like but doesn't really know beyond the form what it is. It writes out answers it "knows" appear correct, not answers that it "knows" has the correct facts. As is. ChatGPT is bad at this stuff. Good RPG games should be consistent.

In a sense, its certainly true that ChatGPT has limits, exactly how much of an internal model the training has produced is the subject of a lot of debate and study. Probably its biggest limitation at the moment is that it can only retain between 4k-8k tokens (being roughly equivalent to words). So longer narratives quickly exceed its ability to keep track of, there is some interesting research work that claims to have raised this limit into the millions. Would be very interesting to see how much better an LLM able to track that many tokens would be.

LLMs are a huge step forward and satisfy many of the conditions of intelligence, with the rate the research is progressing I don't think it'll be long before these systems revolutionise the way games are built and run. Along with basically everything else...

[Juicy Pork Chop]

embarrassing lmao

[Capo Rip]

Originally posted by Vel'Rathis:

Chat GPT is not true AI, its developer Moderated in response rather than a true LLM. Unfortunately that means it has the same political bias as its San Francisco based developers.

ChatGPT is a LLM. Yes it's responses are post processed to be friendly. They are also post processed to be actually responsive and to use the chat interface itself. LLMs and NLP are forms of AI. When you say true AI I suppose you mean generalized AI. No one has got that. LLMs are just chatbots that can be trained on a large amount of data because the data doesn't have to be labeled and tagged. That's the innovation. I get that you're mad that it won't parrot the racsist garbage the LLM has in it's internet based training content. That doesn't make it not true AI or not an LLM. You just don't know what the terms mean.

[grab these bandages]

Originally posted by WeirdWizardDave:

Originally posted by Rex:

No. I'm against ai in all forms.

You're really going to hate the future then.

Yes, already

[WeirdWizardDave]

Originally posted by grab these bandages:

Originally posted by WeirdWizardDave:

You're really going to hate the future then.

Yes, already

Cheer up, we'll probably destroy the planets ability to sustain life before the AI actually takes over.

[Capo Rip]

Originally posted by WeirdWizardDave:

Originally posted by Capo Rip:

You should play with it some more to see it's limitations. The current iteration of ChatGPT doesn't know what facts are and doesn't build permanent or consistent narrative. It knows what a DM and PC player interaction looks like but doesn't really know beyond the form what it is. It writes out answers it "knows" appear correct, not answers that it "knows" has the correct facts. As is. ChatGPT is bad at this stuff. Good RPG games should be consistent.

In a sense, its certainly true that ChatGPT has limits, exactly how much of an internal model the training has produced is the subject of a lot of debate and study. Probably its biggest limitation at the moment is that it can only retain between 4k-8k tokens (being roughly equivalent to words). So longer narratives quickly exceed its ability to keep track of, there is some interesting research work that claims to have raised this limit into the millions. Would be very interesting to see how much better an LLM able to track that many tokens would be.

LLMs are a huge step forward and satisfy many of the conditions of intelligence, with the rate the research is progressing I don't think it'll be long before these systems revolutionise the way games are built and run. Along with basically everything else...

They satisify the surface conditions for intelligence but the very method they use to aquire all of that training data--it uses untagged data--is a real limitation on the "trustworthiness" of it's results. Try some non trivial tests with the free ChatGPT using a topic you know something about.

Obviously this is a big step forward for Chatbots and would likely pass the Turing test for many short, casual exchanges, but right now it would make a bad DM. It does make a great DMs helper. Overcoming the input limit and allowing it to retain "facts" from previous answers will go a long way to bridging the current gap

[WeirdWizardDave]

Originally posted by Capo Rip:

They satisify the surface conditions for intelligence but the very method they use to aquire all of that training data--it uses untagged data--is a real limitation on the "trustworthiness" of it's results. Try some non trivial tests with the free ChatGPT using a topic you know something about.

Obviously this is a big step forward for Chatbots and would likely pass the Turing test for many short, casual exchanges, but right now it would make a bad DM. It does make a great DMs helper. Overcoming the input limit and allowing it to retain "facts" from previous answers will go a long way to bridging the current gap

Sure but trustworthiness isn't a condition of intelligence. ChatGPT like all LLMs can make stuff up that isn't true. But the fact it can make up stuff that is plausible but false, if anything is demonstrator of just how close to true intelligence it is. Calling them chatbots is a bit of a misnomer, I mean people do but ChatGPT has as much in common with the chatBots of the the 00s as Baldurs Gate 3 has in common with a chessboard.

Trustworthyness is an important quality in a useful tool though, I'll grant you that. ChatGPT is already orders of magnitude better at this than LLMs were a couple of years ago. But there is still plenty of room for improvement.

(edit) The Turin test was passed long, long ago.