LLM replaces users in writing automation

[celest fjord]

Since LLM knows my entity and how to write automation, can we directly add the automation code generated by LLM to HA's automation list? Reduce user workload. This is just an idea, I don't know if it's feasible.

[sharp zodiac]

Who says a LLM knows how to write an automation?

[proven vessel]

It’s feasible and has been done in custom integration but it’s in a very experiment state. Don’t expect to write extremely complex automations with it.

Check out the extended open ai custom integration. They were (last time I checked ) providing some info about how to do just that

I would not rely on it to run my house.

[untold crescent]

Don't think it will be flawless. Even GPT-4o cannot cross-check itself, and lies confidently. The output always should be checked by human.

[celest fjord]

@untold crescent@proven vessel Self written automation also requires inspection and testing, even if LLM often lies, this can be corrected through inspection. At least it is feasible.
Devices like self discovery can be discovered by LLM and added on our behalf, actively prompting us on how to collaborate with existing devices to generate automation. But we cannot deny its convenience.

[untold crescent]

Yeah, but writing that to the automations right away means it can run without human inspection.

[celest fjord]

seems that someone is already working hard for this https://github.com/ITSpecialist111/ai_automation_suggester

[sharp zodiac]

I get a 404

[celest fjord]

@sharp zodiac 😅