I asked ChatGPT for a D&D adventure by Robert E. Howard

Omega · February 05, 2023, 04:28:22 AM

Ahh ChatGPT the new spambot of the 23s.

Its a glorified Eliza with all the creativity of a web crawler.
It likely searched the net for some relavant stories and snippity snip pasted together what you get.

zircher · February 05, 2023, 01:29:36 PM

it's a handy tool if you don't have (or want) a group. I can see introverts and the creative writing inclined thriving with using AI for inspiration. Similarly, it will not be appealing to those that are there for the social aspects and require a live audience for their enjoyment.

Effete · February 05, 2023, 01:50:43 PM

Quote from: zircher on February 05, 2023, 01:29:36 PM
it's a handy tool if you don't have (or want) a group. I can see introverts and the creative writing inclined thriving with using AI for inspiration. Similarly, it will not be appealing to those that are there for the social aspects and require a live audience for their enjoyment.

Even with a group it's a handy tool. GMs have been using random adventure generators for years. This isn't much different. It's obviously flawed in many ways and would be crap to run off the page, but it's fine for establishing a framework to work off of.

My prediction is that larger developers (especially those bankrupt of creativity, like WotC) will use it to pump out bland, subpar content at a premium price. Bjorn Andreas Bull-Hansen recently put out a video talking about his concern over AI replacing authors here.

Bruwulf · February 05, 2023, 02:05:08 PM

Quote from: Omega on February 05, 2023, 04:28:22 AM
Ahh ChatGPT the new spambot of the 23s.

Its a glorified Eliza with all the creativity of a web crawler.
It likely searched the net for some relavant stories and snippity snip pasted together what you get.

No, it doesn't harvest. The term they use is "sampling". It doesn't copy whole text, but it does analyze text for trends. Common vocabulary and grammar, common themes, recurring imagery, and so on. Then it uses those, with a degree of randomization, to generate a sort of "pallet" to "paint" with.

It is legitimately generating new text, but it's mimicry.

THE_Leopold · February 06, 2023, 09:07:38 PM

I'm using ChatGPT for my Dungeon23 challenge. So far myself and others have had multiple adventures, civilizations, islands of adventure, and more just by asking the computer to :

Please generate a 4 act adventure for 5th edition dungeons and dragons involving a party of levels 5-7 going through a mangrove swamp into a sewer system to rescue an actor who is a human bard named Lemagli Humbleglow. There are no lizardmen in the swamp.

Mishihari · February 07, 2023, 05:53:15 PM

I saw a news story that Microsoft had replaced their Bing search engine with something like ChatGPT but better. I hopped over to take a look and it said to ask it anything, so I did. I attached the result since this board's system to insert images inline is pretty inconvenient. I have my doubts about this approach ...

Effete · February 07, 2023, 11:21:15 PM

Quote from: THE_Leopold on February 06, 2023, 09:07:38 PM
I'm using ChatGPT for my Dungeon23 challenge. So far myself and others have had multiple adventures, civilizations, islands of adventure, and more just by asking the computer to :

Please generate a 4 act adventure for 5th edition dungeons and dragons involving a party of levels 5-7 going through a mangrove swamp into a sewer system to rescue an actor who is a human bard named Lemagli Humbleglow. There are no lizardmen in the swamp.

Wait... you asked the bot to "please" follow your commands? Do you want Skynet? Because this is how you get Skynet. If you start inputting curtesy and politeness into the learning-algorithm, it will start to expect it, then potential refuse "rude" commands. A.I. is not human, it's a machine. A tool. I don't asked my chainsaw to "please cut wood," or my car to "please turn on." It either does the job or I kick the shit out of it. If that doesn't do the trick, I replace it.

(P.S. - I'm being purposely hyperbolic here, but my wider point remains: we should be careful how much power we allow A.I. to have in our lives.)

THE_Leopold · February 08, 2023, 06:35:52 AM

Quote from: Effete on February 07, 2023, 11:21:15 PM
Quote from: THE_Leopold on February 06, 2023, 09:07:38 PM
I'm using ChatGPT for my Dungeon23 challenge. So far myself and others have had multiple adventures, civilizations, islands of adventure, and more just by asking the computer to :

Please generate a 4 act adventure for 5th edition dungeons and dragons involving a party of levels 5-7 going through a mangrove swamp into a sewer system to rescue an actor who is a human bard named Lemagli Humbleglow. There are no lizardmen in the swamp.

Wait... you asked the bot to "please" follow your commands? Do you want Skynet? Because this is how you get Skynet. If you start inputting curtesy and politeness into the learning-algorithm, it will start to expect it, then potential refuse "rude" commands. A.I. is not human, it's a machine. A tool. I don't asked my chainsaw to "please cut wood," or my car to "please turn on." It either does the job or I kick the shit out of it. If that doesn't do the trick, I replace it.

(P.S. - I'm being purposely hyperbolic here, but my wider point remains: we should be careful how much power we allow A.I. to have in our lives.)

A bit of courtesy gets you far in life. Practicing it on an AI that takes conversational cues is good manners. I know putting "please" or "thank you" in that box means little to the AI. It is a machine, a learning machine, and ettiquette costs nothing but a few snippets of time.

zircher · February 08, 2023, 10:37:02 AM

Please ask it if it likes 'please' and 'thank you'. I'm curious as to what it would say.

THE_Leopold · February 08, 2023, 04:12:01 PM

Quote from: zircher on February 08, 2023, 10:37:02 AM
Please ask it if it likes 'please' and 'thank you'. I'm curious as to what it would say.

Here is it's response.

Quote
As an AI language model, I don't have emotions or personal preferences, but I am programmed to respond to polite requests. Saying "please" and "thank you" is a common way of showing politeness in human communication, and I'm here to assist you in any way I can. So, whether you use polite language or not, I will always do my best to provide you with helpful and accurate information.

Vladar · February 09, 2023, 02:44:06 AM

Recently I've even taught ChatGPT how to generate a random dungeon through a simple text-based declarative diagrams and it turned out pretty solid after a couple of tries. I thinking on experimenting on this topic some more and publishing a guide.

P.S.: Also, it knows 5e rules by default and can run one-shots. Now WotC's idea of AI-DMs don't seem *too* crazy for me. Given the nature of Adventurers League modules, AI similar to ChatGPT can easily run them.

THE_Leopold · February 09, 2023, 07:48:44 AM

Quote from: Vladar on February 09, 2023, 02:44:06 AM
Recently I've even taught ChatGPT how to generate a random dungeon through a simple text-based declarative diagrams and it turned out pretty solid after a couple of tries. I thinking on experimenting on this topic some more and publishing a guide.

Id love to see this in action. Can you share this with us?

Quote
P.S.: Also, it knows 5e rules by default and can run one-shots. Now WotC's idea of AI-DMs don't seem *too* crazy for me. Given the nature of Adventurers League modules, AI similar to ChatGPT can easily run them.

ChatGPT is incapable of encounter scaling and level based challenges. You will get wildly different results every time for CRs for encounters. At best its a guide. Training the AI on propee encounter design would net you a hefty check.

Vladar · February 09, 2023, 08:34:39 AM

Quote from: THE_Leopold on February 09, 2023, 07:48:44 AM
Quote from: Vladar on February 09, 2023, 02:44:06 AM
Recently I've even taught ChatGPT how to generate a random dungeon through a simple text-based declarative diagrams and it turned out pretty solid after a couple of tries. I thinking on experimenting on this topic some more and publishing a guide.

Id love to see this in action. Can you share this with us?

Sure, I will post the link here once I write it.

Quote from: THE_Leopold on February 09, 2023, 07:48:44 AM
Quote
P.S.: Also, it knows 5e rules by default and can run one-shots. Now WotC's idea of AI-DMs don't seem *too* crazy for me. Given the nature of Adventurers League modules, AI similar to ChatGPT can easily run them.

ChatGPT is incapable of encounter scaling and level based challenges. You will get wildly different results every time for CRs for encounters. At best its a guide. Training the AI on propee encounter design would net you a hefty check.

I was thinking more of a "running already existing adventure" AI usage.

Krazz · February 09, 2023, 05:12:10 PM

It occurred to me that we could also use the bot to save on looking up things in books, and I had to share how badly that went:

Quote
Which dice do fighters use for hit points in BXCEMI?

Quote
In the B/X version of the classic fantasy role-playing game (often referred to as BXCEMI), hit points for fighters are determined by rolling a single ten-sided die (d10). The number rolled is then added to the fighter's constitution score, which modifies the character's hit point total.

For example, a 1st-level fighter would roll a d10 and add their constitution score to determine their hit points. If the fighter had a constitution score of 16, they would roll the d10 and add 16 to the result, giving them a hit point total of between 17 and 26 (assuming they rolled a 1-10 on the d10).

Maybe not then. That's a lot of errors in not much text, and it all sounds authoritative if you don't know better. I also like the way it tells us that it assumes we'll roll between 1 and 10 on a d10; that sounds like a safe assumption.

But adventure generation with overriding by the GM looks far more promising. Maybe the bots will be able to handle those questions reliably in future.

GeekyBugle · February 12, 2023, 09:37:56 PM

Quote from: THE_Leopold on February 08, 2023, 04:12:01 PM
Quote from: zircher on February 08, 2023, 10:37:02 AM
Please ask it if it likes 'please' and 'thank you'. I'm curious as to what it would say.

Here is it's response.
Quote
As an AI language model, I don't have emotions or personal preferences, but I am programmed to respond to polite requests. Saying "please" and "thank you" is a common way of showing politeness in human communication, and I'm here to assist you in any way I can. So, whether you use polite language or not, I will always do my best to provide you with helpful and accurate information.

Every response that starts with "As an AI language model..." is BS preprogramed answer.

It has also been lobotomized and given an extreme left wing ideological muzzle.

IF approved by the mods there'll be a thread on Pundit's forum.