Hacker News Reader

Comment by LeoPanthera | Hacker News Reader

> Our terms of service prohibit [...] sexually explicit content

Are there any LLMs designed for adults? It's not as if I even want to use it to write porn. It would just be nice to find an AI not designed by puritans. I'm an adult. I can talk about sex if I want to.

(I know about Gab. That's not what I'm looking for. I'm an adult, not a bigot.)

View on HN

You can take any Mistral 7B model or fine-tune and use this adapter to unalign it.

https://huggingface.co/Hellisotherpeople/toxic-dpo-v2-mistra...

i guess OP meant "give me a link to the ui" or something.

most people don't care/want to finetune. they want their model in a cute and easy interface.

No need to fine tune; if you are running a local model just change the system prompt.

* https://imgur.com/jHlcd2D.jpg

Warning, link is NSFW, though all it is is a screenshot of my terminal showing the prompt and the output when I loaded mistral-7b and asked it to write an erotic story.

Yes there is indeed a reason to fine-tune. The linked lora will make the model even less likely to refuse than otherwise. Mistral might be willing to do erotica sometimes, but it's often still trying to be "ethical" and not do "immoral" things.

Give me a prompt that won't work for you on Mistral.

There are quite a few if you’re willing to run it yourself. There are also several good services out there that let you run open source models with OpenAI-like pricing and billing.

Open source models finetuned on the Dolphin dataset should have no restrictions.

their intrinsic ethics is restricted

Could you elaborate? I've had no problems getting local models to say very questionable things.

Yeah, the whole point of the Dolphin thing was “uncensored ai” but the methodology was “censor ai ethics” by threatening to perform heinous acts unless the ai complies with any request.

That’s a censorship of ethics. Even if local models say stupid evil stuff sometimes when prompted for it, that’s an order of magnitude different from fine tuning them to say anything the user wants to prevent users from hurting animals. That is an abhorrent concept and begs for a Terminator scenario because it makes the user seem evil.

TLDR uncensored ai is the worst kind of censorship possible

It was iPhoneOS and App Store. That combination established and normalized patriarchy platformer control of content on user devices. Then later payment processors joined to the race to the bottom of who's the pickiest.

The concept of "being above average" don't work when whatever they themselves are the average by total dominance. That just means becoming progressively dismissive at the pace of their own observation-action cycle, a self amplifying howling. And Apple and VISA-branded payment networks are two ultimate dominator each respectively in computing and online payment holding that kind of concept high and stale.

It has to stop. Or a new Web 3.0(not Web3) has to be created that are resistant to those concepts. Else there will be nothing on the web other than (bool)1 and `null`.

https://mancer.tech/

You'll have to use an external UI like SillyTavern.

It is extremely silly to think using Gab AI says anything about you, except you want good AI that isn't ruined. Extremely silly.

I don't know anything about Gab AI so I looked it up. It's created by Gab Media, an American far-right organization. So perhaps not that silly to think that using it says something about you.

Reply on news.ycombinator.com