Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Skynet]

On September 30th 2021, Latitude changed its policy regarding censorship to something that is a bit more reasonable than what they had before. Specifically:

- There will be no moderation of unpublished single-player content. Instead, technological barriers (the "content filter") will exist to try to limit a small subset of generated speech. No consequences for users who trigger said filter...or find a way to bypass them.
- The stories will be encrypted to protect the company from looking at them, thereby reducing the possibility of censorship.
- There will be guidelines that will be applied to published single-player content and published scenarios. These guidelines do not apply to unpublished single-player content though (which, I think, is fair).
- OpenAI can, of course, impose their own filters and restrictions - if so, Latitude will allow you to switch to a custom-made AI algorithm (later named "Dragon-21") that do not have those filters and restrictions. (Griffin also does not have any of those restrictions as well...so if you don't have the premium version, you can still use Griffin instead.)

On December 31st 2021, the company announced that most people have switched over to Dragon-21, meaning that they no longer get a discount for using OpenAI's GPT-3...which means, they're no longer going to offer GPT-3 and stick with Dragon-21.

With the higher traffic we had a volume-based discount that we will no longer have since most users transitioned away. Because of that we would now pay the normal token price which is 12 cents / 1000 tokens (which is around 1 action) for a fine-tuned model. That would mean every 100 actions would cost $12.

Given this change and that most users can’t have a good experience with the new filters, Dragon-OA isn’t going to be offered starting in the new year.

...or, in conclusion...

R.I.P game...

The game has been raised from the dead.
---
Meanwhile, OpenAI has opened up a public beta, and I've been playing around it. The outputs are super-cool, much faster than AI Dungeon, and pretty cheap too. But there are a lot of restrictions involved when using it, and the company can take away your access if you don't follow their guidelines. So I have to decide what I want - the quasi-freedom that AI Dungeon promises or the performance and affordability of OpenAI. It's a hard choice to make, but I lean towards quasi-freedom.
---
EDIT: Just want to add some words about pricing. AI Dungeon would need to pay $0.12/1000 tokens to use a fine-tuned model from OpenAI, but if you instead use OpenAI's normal model directly, you only need to pay $0.06/1000 tokens. I just used OpenAI's model to create a character profile, which "cost" me $0.03 ("426 prompt + 30 completion = 456 tokens"). Note that I use "cost" in quotation marks, because I'm currently using OpenAI's free quota, just to try their algorithm out and compare it to AI Dungeon.

Since I don't use AI Dungeon often, it's possible that I could save money switching to OpenAI...but that does require me to comply with their regulations, and honestly, quasi-freedom starts to sounds a lot more appealing when you consider the sheer number of regulations they expect me to follow.

[Muz]

But there are a lot of restrictions involved when using it, and the company can take away your access if you don't follow their guidelines.

I've been using OpenAI for half a year. They're fine as long as you don't do damage with it and follow their guidelines. They do allow you to do uhh, experimental stuff with their AI, just don't share the bad results with everyone or do it at scale. The AI is also a lot better at these things than most humans.

I've actually been using it for some solo RP stuff myself. AI is generally bad at maintaining state of the world, or conflict resolution, but it's perfect for solo tabletop. Might start a thread on that once I figure it out.

[tnc]

My first game a woman was looking for her son and I went out of the store we were at and her son was right there. I brought her son inside and the kid went "Mama! Mama!" but when I pointed out her son to her the game said she looked at the kid but she still kept asking me if I could help her find her son. Another game there was an evil Snow Queen who brought chaos to the world. I found her and made sweet love to her and the game said there was nothing left in the world except me and her so I think she just destroyed everything except me.

[Strongpoint]

I switched to Kobold AI. It isn't as powerful as Novel AI or AI dungeon and requires Google colab ( and its mercy) to function but I very much prefer reasonable opensource solutions

[Cthulhu]

Novel AI seems pretty good, I switched over cause of AI Dungeon's censorship (i'm not tryna do bad shit but that doesn't mean I want people looking at my non-bad shit)

How is Dragon-21 compared to GPT3?

[Broseph Stalin]

I came back after they implemented encryption and stopped moderating unpublished stuff. New dragon is dumber than old dragon but it's better than the early weird google document version which is the one I declared a text-base holodeck.