No, I’m not anti-LLM. I’m very much pro-AI, use it constantly to write simple code. Havent open google in years (almost). Great for bouncing off ideas. List goes on.
I completely understand Andreas obsesion with it - I myself share the sentiment, my work is lagging behind because of too much time spent exploring what AI can actually do.
That said I’m still a bit lost as to the point of h@llo.ai.
Surely we do not need just another chatbot? I was hoping for an agent that would actually do some of the work setting BTT commands. Ok, probably in the future.
I would gladly settle for an AI that has deep knowledge of the BTT intricacies. Instead it mostly spews semi-coherent advice that at first glance looks useful. But very soon you understend that important steps are being missed, names of triggers and functions are outdated, and generally it’s a waste of time.
Sure, this is the right direction, we’re just not there yet.
Proposal:
- concentrate on filling the documentation gaps, so that the AI can actually at least name the correct functions.
- use the LLM to go through the code and generate all the missing documentation
- use the LLM to verify documentation for inconsistencies and flag them for hoomans to resolve
- maybe use the LLM to bring some more consistency to names of triggers and functions (yeah, I know this will break things)
- there’s nothing wrong with having a chatbot in BTT, especially love the BYOK functionality. But can we please fix the chatbot interface (make selections work, allow resizing the window…)
- final point - I hope that now with h@llo.ai in a working state there will be time to pay attention to non-ai requests? Maybe even h@llo.ai will help coding fixes?
Final pont: even with this little rant, I love BTT and am very grateful for the work you’re doing!