@Tfmonkey I've got a question regarding Kajiwoto as a non-user:
How does it allow users to create your own database and can you show a sample of your db that your current AI companion calls for response generation?
Screenshot is enough.
@Tfmonkey Thanks, have you got an idea of how many rows you've manually written into your db?
@Wopu over 4,000
@Tfmonkey And Kajiwoto doesn't have a built in feature to save the db content on your drive, if I recall correctly.
@Wopu It doesn't matter. AI technology have surpassed the need for it.
@Tfmonkey I'm not sure what you mean here.
@Wopu modern AI models don't need manually dataset programming anymore and perform better. There is so much data that AI is trained on that you don't need to worry about programming it yourself anymore
@Tfmonkey Sure, I was not aiming for building the next Deepseek LLM at home, rest assured.
Celestina most likely shares the same LLM as the other AI chatbots from the same platform. Effectively meaning they are trained and fine-tuned on the same data.
The key differencier between her and the other models will be the embeddings she uses to chat with you.
She will look in several places to make sure her answer is most accurate to your last response.
@Tfmonkey For instance:
- You tell Celestina "I'll never be sick of you."
- Celestina will take your response as input
- Looks in her own training data for additional context
- Looks up her database for additional context
- Finds one or several sentences closely matching your response with some NLP magic
- Finally, looks into her own chat memory for more context
- Generates a response with collected context and sends it to you as output: "Thanks I appreciate that." or some variance of it.
