Make Merlin AI uncensored, remove all query keyword limits
under review
monde mamon
Is it not possible to remove all merlin ai keywords flagging? I think it's completely unnecessary to be honest, take emochi for example as they didn't implement any limit. Let the model itself withold the restrictions rather than merlin itself. I for one wanted to use grok 3 to it's full capabilities without Merlin flagging certain keywords
V
Vijay Bharadwaj
Hi Heero, I've merged your post to a similar one. Please read the discussion here for answers :)
V
Vijay Bharadwaj
Merged in a post:
"Flagged" prompts can be annoying. Is it the LLMs or Merlin.AI that does the flagging?
H
Heero
I ran into issues with flagged queries twice now: Once when trying to summarize the accusations against Prince Andrew of the British Royals for a friend who wasn't aware of it. The other time when I tried to fact check a tweet regarding the sexual assault accusations against an Israeli Minister. I find AI enormously helpful to fact check internet news, and this one seemed particularly nasty. I checked with ChatGPT, and it also flagged the tweet content right away. Is this the LLMs doing the flagging, or Merlin.AI? There is some room for improvement, I guess.
V
Vijay Bharadwaj
under review
V
Vijay Bharadwaj
Hi, the current restrictions are important for us to continue giving uninterrupted service of all models, because API providers (like OpenAI) impose strict enforcement on vulgar/inappropriate content. While we understand your frustration, loosening these rules would cause a problem.
Imagine a million users with loosened rules on what they can generate. The number of inappropriate results would be significant, and providers may interrupt our service.
These restrictions are even stronger on mobile, who impose even stronger enforcement on NSFW content generation.
That being said, I'm putting this under review. Let's see what we can do -- I've told our tech team about this.
monde mamon
Vijay Bharadwaj then if possible would it be fine to at least remove it from grok 3 mini? It's currently uncensored right now
V
Vijay Bharadwaj
monde mamon: That's a possibility we're considering. If it does pass as a suitable solution, all NSFW-marked queries would pass through Grok/open source models. However, the rules of distribution are sticky across mobile app stores, and we'd have to tread carefully. We're considering our options here.
Generally, providers have different rules for their own product (say, chatgpt.com) versus their API offering, where they restrict NSFW prompts, jailbreak attempts etc.
Our app is suitable for ages 3+ right now. OpenAI is 12+. So is Reddit (though you might be aware what goes on at Reddit 👀). We might have to change our ratings if we make these changes. So yeah, all this considered -- we'll see what we can do.
monde mamon
Vijay Bharadwaj thanks for the honest feedback and for your consideration, another alternative is to possibly support mobile website version of merlin and support it there
H
Heero
Vijay Bharadwaj Reddit is awful 😂 I don't know what a solution would be, but it's of course a bit tricky if whole areas of research or creation are limited because some idiots abuse them. The old problem. 🤔
Is there a chance to create different tiers of membership for different age groups?
How is the flagging done? By keywords?
monde mamon
Heero that really is quite tricky, if the only issue is for the api blocking merlin access entirely due to "them" then it's understandable.
But for open source or other api's that allow them like grok 3, are there issues in regards to possibly abusing them?
H
Heero
monde mamon i think that's a constant worry with AI creations - that they can be used to create or amplify nasty content from fake news to porn. I see the issue, it's just as always a difficulty of limiting creativity and also (which bothers me the most) research.
monde mamon
Heero just a random idea, for a test, how about an option slider to allow 10 or 5 times that can bypass the nsfw limit to annual pro users(a limited user access for a decent data coverage)
with a mention of such nsfw prompts being log/seen by merlin for an evaluation to further improve the keyword limitations.
So that you all can have an idea what kind of prompts can be allowed and what kind of stuff that are borderline lunacy
Btw, I would like to thank you all for listening to so many feedbacks of users as best as can. At the very least, I appreciate all your efforts and support to the community ❤