Hi!
I am getting late into CSAI, maybe this was discussed before.
Would it make sense to have an ability to point CSAI to training docs? I guess that generating text in a text element can take the page context, but can it take the website context?
Having a plaintext training doc with the website details, its goals, tone, voice, examples… would be super useful.
Indeed. Hopefully this can be done with their own upcoming AI provider, though based on Charlie’s response here it may not initially include such context training.
There is a global setting called “System Prompt” where you could point CSAI to docs or just paste in docs here. We could read the site name and site tagline, however I’m not sure how much info that would supply to AI. Does this help?
This is perfect! Especially if links to docs can be read. I will play with it to see what is the optimal balance between the amount of info and token costs, among the models.
But even such a small local field is great to teach LLM not to behave like LLM. 
