Demod for DeepSeek

[nano-snooter-booper]

Deepseek R1 appears to blow OpenAI's current model(s) out of the water, at least in value (actually free and open) as well as being far less censored currently. The problem is, they seemingly added the exact same automod on their site as chatgpt put in. It didn't seem to initially have it when I first used their model on https://chat.deepseek.com/, but now if it sees a word it doesn't like, an entire message goes away when it's done writing and it even attempts to gaslight you "Sorry, that's beyond my current scope. Let’s talk about something else." And even if you ask it to write again, unless you say exactly what you're wanting to do it seems to force the memory of your current conversation out the window, at least of your last ask. What I'm betting on here is this is quite similar in form to the one on the chatgpt site. Is it possible for this script to be remade for deepseek, and are you considering it? Yes, I know this might not be the best page to ask for this, I just don't see a message function on github and dming you feels more demanding, I'm just hoping, not begging.

[nano-snooter-booper]

Dug through in inspect element a bit, not an expert but it looks like when the mod event triggers it does overwrite the text with 'Sorry, that's beyond my current scope. Let’s talk about something else.' or something close, and doesn't keep the message stored. As long as that event is interfered with, that should be preventable, right?

[nano-snooter-booper]

If you haven't tried deepseek out yet, it really is great. It's writing ability is on par with or better than chat, and it rejects little outright, it's just the pesky message wiping censor that is the big problem. Without it there deepseek would be back to being the best there is right now.

[4as]

Well, if we want to talk about better alternatives openrouter.ai has multiple free models without any dedicated filtering. You can also use Google's ai studio, which has options to remove filtering for its Gemini models.
Personally, I'm so tired of dealing with refusals I just run AI models on my own PC. Even with 6GB VRAM you can get some interesting results.
Which is why I've given up on writing scripts like DeMod for other AIs - it feels pointless.

Still, I'm going to leave this suggestion up for now. Maybe I'll take a look if more people request it.

[nano-snooter-booper]

Thanks for considering it, personally I don't have enough VRAM at my disposal to locally run anything worth a darn, and just stuck with pre 4000 series cards, mobile ones. A model like Deepseek coming around really shook things up, though Cohere wasn't too bad in its stead and employs none of the heavy handed censorship. Thing is I just hadn't found an AI that could truly, properly take on Chatgpt on the creative writing level until Deepseek R1. It blew me away, and they've already begun the race to ruin it. I see what you mean by that with how fast chat became nigh draconian. I'd suggest trying Deepseek to see what I mean. People aren't overhyping it. Doesn't require as much to sign up as Chat and I'd assume Google's not great on that either. Deepseek is also open source as mentioned, though to run it at a level anywhere near what you get through their site you'd probably have to run their 70B model - they do have a real small one that should be ok on 6gb vram, haven't tried it. Unlike OpenAi, they aren't closed. Their github - https://github.com/deepseek-ai/DeepSeek-R1

[nano-snooter-booper]

And thanks for the suggestions, hadn't known about Openrouter. They've actually got Deepseek R1 70B on there already. Gonna try it if it's reasonably priced to run and report back. Might not be necessary to put any work in on a demod after all if anybody else asks.

[nano-snooter-booper]

Using Deepseek v3 on there, it's much more likely I've been chatting with it on Deepseeks site, it is what they adverise mainly, R1 has some button you can toggle. Ooenrouter seems a little slow so far but using their v3 writing wise it seema to do about as good! And I'm assuming there aren't any mod tricks like chat pulls. Thanks again.

[nano-snooter-booper]

Changing providers sped things up. Thank you so much for bringing this to my attention. This should be able to replace using their demo chat site before that becomes unusable from censorship, and I don't mind paying when people are willing to host models like these.

[venividicode]

Well, if we want to talk about better alternatives openrouter.ai has multiple free models without any dedicated filtering. You can also use Google's ai studio, which has options to remove filtering for its Gemini models. Personally, I'm so tired of dealing with refusals I just run AI models on my own PC. Even with 6GB VRAM you can get some interesting results. Which is why I've given up on writing scripts like DeMod for other AIs - it feels pointless.

Still, I'm going to leave this suggestion up for now. Maybe I'll take a look if more people request it.

Might I ask, how does one run an AI model locally, and which ones?

[4as]

Might I ask, how does one run an AI model locally, and which ones?

It should be relatively easy to find guides online. I've also written one.

[quangnh89]

Might I ask, how does one run an AI model locally, and which ones?

I use ollama (https://ollama.com/) (back-end) and Open WebUI (https://github.com/open-webui/open-webui) (front-end)

You also use Koboldcpp (back-end) and SillyTavern (front-end).