Problem solved
Thank you all
Since
Post must be at least 100 characters , that’s why I’m still typing like a dumb xD
this session has some instructions on how to replicate the behavior, perhaps.
1 Like
No, this is complete nonsense and your GPT is hallucinating.
100% not. OpenAI says its models can go up to 1m in context length… but realistically this is only for needle-in-haystack tasks. Anything else and its actual context size is much lower, and attempting to go above that will result in what you’re seeing here.
I would thumbs-down any more hallucinations you’re seeing like this.
I also recommend starting a new thread whenever the topic changes. You can add traits to ChatGPT in your settings. I strongly recommend against reusing the same thread for everything, because it usually results in the AI going whackadoo.
no need to,
your prompt literally asks for the behavior
“• Characteristics: 羞耻 trigger, ego defense, language”
It’s also just a very loaded prompt with a lot of different behaviors being asked for. I see a few things that might create issues:
- Instructs the model to hallucinate. It may be better to instruct it to roleplay and bring in its character’s past events.
- The prompt constantly defines new terms, and they often don’t make sense either. This might be overwhelming it, and causing it to invent new concepts on its own too. You may have better results just telling it what it needs to do. Avoid being fancy.
- Prompting with instructions in any language other than English usually degrades performance.
This is a very large prompt defining a large number of behaviors. At that point I’d just make a fine-tuning dataset based on outputs you’ve seen and liked, and maybe some handwritten ones. Not sure how okay the NSFW stuff may be as far as policy goes. You likely will have problems fine-tuning with any more than a few NSFW examples.
1 Like
Tbh they should turn off the filters for a month, see what users do with the freedom to create NSFW and from there see what they prohibit and what they allow, as a way of knowing how to implement NSFW without it falling into illegality or shady situations. It would be the fastest way for them to know how to implement a NSFW system, many say it would be good to add age filters and that to activate adult mode there is age confirmation like in Tinder or Bumble and others say that it should only be in plus mode, but not all adults can afford to pay 20 dollars a month and less when you don’t earn in dollars.
at this point the guy should just train his own personal Ai if he’s that demanding of it’s character…
~just sayin~