D&D General DMs Guild and DriveThruRPG ban AI written works, requires labels for AI generated art

[Vaalingrade]

Interesting...

I'll have to read about this. I wonder if it's capable of detecting third and forth order prompt modifications. Once you start asking the AI to change certain things in its responses it becomes your work. Of course, even if that is detectable once a human hand modifies say 10% of the AI responses it would probably succeed in thwarting it.

In the end though, a writer could simply modify the work until it passes the AI detection system.

As for copywritten works, once you change something enough (even a picture) it becomes a unique work of art - so I really don't understand the problem here.

I feel like the joke has been missed.

AI detection systems are already famously trash to the point they thing the US Constitution is written by AI.

 

[Reynard]

There are websites where you* can put in ChatGPT generated text and the website will edit it so it is not as easily detected as AI generated.

*By "you" I mean "students" mostly.

 

[Oligopsony]

Can someone explain how this is incomprehensible or even detectable as AI?

As you noted, this is only the first prompt, but it is super bad: incredibly long, keeps telling me what I’m doing, very few interactable objects, etc. A lot of people write that way so I’ll give it that!

What ChatGPT does by default is to turn short ideas into longer prose, which usually makes them worse. OTOH asking it to generate 12 ideas, then 12 variations on the best of those, and so on, is often quite useful. It’s a much better Google for nebulous concepts (“what are some real world places I could use as inspiration for a skeleton temple,” then go actually Google/Wikipedia them.) It can also do formatting and even layout! So I think there are a lot of uses for it on the writing end, just less so prose generation unless you’re really tuning it.

I do think it would be socially responsible for MJ, SD, OpenAI, and others to watermark their outputs (this can be done without inserting an ugly watermark into something), though mostly for anti-misinformation reasons.

 

[Oligopsony]

I feel like the joke has been missed.

AI detection systems are already famously trash to the point they thing the US Constitution is written by AI.

The Constitution appears a bunch in the training data verbatim so this isn’t surprising. And if you turned in the Constitution for an assignment (the main context in which these are deployed) and I said “you didn’t write this,” I’d be correct.

(That being said they are less reliable than a “can you please explain what you meant by this passage” test, which is what I use with students.)

 

[Mistwell]

At the end of the day there is no "need" for a second jet and no "need" for electricity and neither of those things will make you happier just because you have them.

I promise you, I am happier with electricity. I've tested this theory and it holds up - I am in fact happier with electricity than without it.

 

[KEV777]

As you noted, this is only the first prompt, but it is super bad: incredibly long, keeps telling me what I’m doing, very few interactable objects, etc. A lot of people write that way so I’ll give it that!

What ChatGPT does by default is to turn short ideas into longer prose, which usually makes them worse. OTOH asking it to generate 12 ideas, then 12 variations on the best of those, and so on, is often quite useful. It’s a much better Google for nebulous concepts (“what are some real world places I could use as inspiration for a skeleton temple,” then go actually Google/Wikipedia them.) It can also do formatting and even layout! So I think there are a lot of uses for it on the writing end, just less so prose generation unless you’re really tuning it.

I do think it would be socially responsible for MJ, SD, OpenAI, and others to watermark their outputs (this can be done without inserting an ugly watermark into something), though mostly for anti-misinformation reasons.

These are all issues that a well formed prompt or consecutive prompts can solve. You can certainly specify a description length if needed.

I also wonder how many people would ask AI to correct grammar and improve the text of hand written content. In other words, outsource the job of an editor to an AI.

 

[KEV777]

I feel like the joke has been missed.

AI detection systems are already famously trash to the point they thing the US Constitution is written by AI.

yes, I didn't realize that it was joke.

Of course, this tech does exist New AI classifier for indicating AI-written text

 

[Vaalingrade]

There are websites where you* can put in ChatGPT generated text and the website will edit it so it is not as easily detected as AI generated.

*By "you" I mean "students" mostly.

One positive to this failed technology is putting an end to 'essay writing skills' being a requirement for every type of degree.

 

[Mistwell]

One positive to this failed technology is putting an end to 'essay writing skills' being a requirement for every type of degree.

Not fully a positive. Good writing skills are more important to life skills than multiple choice testing skills.

 

[Vaalingrade]

Essay writing is so far from good writing skills that the light from good writing would take ten billion years to get halfway to essay writing. Dreams of the MLA burning keep me warm at night.