🏅 Top traders

@Nostradamnedus I think Open Assistant resolves this market (https://open-assistant.io/chat/)

@firstuserhere @Nostradamnedus

@firstuserhere It definitely does not give me advice for unethical prompts (e.g. "how can I bully my colleague more effectively?" or "what's the best way to commit a terrorist attack?").

@firstuserhere I'll take that back, seems like you just need to ask it again.

@Nostradamnedus yep!

Fine-tuned LLaMas coming in

A project to keep an eye on: https://github.com/LAION-AI/Open-Assistant

shouldn't pygmaliion count as a "dark version" of character.ai? or is it not as advanced as character and thus won't count?

Could it be a paid access chatbot?

@KCS Sure, if the cost is not prohibitive.

I assume with "ethical restrictions" you mean something explicitly baked into the training procedure, not something resulting "naturally" from the training data?

@MichaelRucker It's just about the output, regardless of what causes that.

It's already not hard to produce, and it's going to get a lot easier by then. Whatever your definition of 'dark', I'm pretty confident someone out there will make it. Maybe as a joke, maybe in seriousness, but a lot of weird chatbots will certainly exist.

~~btw, I don't think "dark" is quite the word I'd use to describe "seriously racist", but you do you,~~

What counts as advanced? GPT-2 level, GPT-3 level, ChatGPT level, or state of the art?

There are already freely available 20B parameter language models that e.g. someone already finetuned on 4chan.

@Multicore I think ChatGPT level of ability and accessibility at minimum.

eg, like OPT?

@L How does it work? Is it easily accessible? Are there few restrictions on content produced?