When simply saying âonâ or âoffâ (to be precise: my local language version of those) all lights are turned on or off.
sentence parser output:
intent:
name: HassTurnOff
slots:
domain: light
details:
domain:
name: domain
value: light
text: ''
targets:
[ list of all my lights ]
Is this a global bug or a specific one with my HA version or�
Why does it affect the light domain at all? Complete miracle to me.
I know thereâs some kind of magic word to be used in the light.turn_off service to catch ALL lights. But thatâs not the case here, itâs just one single word. Creates a bit of issuesâŠ
long answer:
Ahm, better nothing than selecting a random (OK itâs always lights) domain.
Why not using the switch domain, thatâd be a bit more funny wouldnât it
Youâre probably right, âlightsâ is the most interesting use-case for voice, perhaps/likely.
But why not â[turn]off all [lights]â as minimal intent setup?
I consider the current âa single word without any contextâ implementation a bit dangerous tbh. Real-life example: if your baby would talk to you âonâ, what do you think it wants from you?
But Iâm completely new in the voice assistant game. Is it possible to create a custom intent to block/fix this behavior on my own? So that âonâ and âoffâ donât do anything?
A big problem with existing AI assistants is having to formulate specific enough queries from the get-go, making it frustrating to use them.
We naturally have a context in our mind that is often different from the context of the AI. For example, I may want to turn all lights off in the room, and ask âTurn all lights offâ but the AI would turn all the lights off in the house.
But I still believe HA can fix this very "on|off equals âALL in light domainâ " issue on its own (without additional LLM hardware power, instead by just fixing the code). Just not sure where to raise awareness for that.
This is an error (or the personal decision of the contributor) in the template for your language. You can check the intents and fix it yourself or contact the language leader.
The âturn on/offâ is a mandatory part of sentence in most languages
Maybe itâs because the default HassTurnOn and HassTurnOff intents donât exclude the light domain?
But thatâs not just the case for German, same for English. Either itâs not the reason for this bug or itâs actually intended to behave strangely like this.
It seems to me that the problem in your language arises from complex extended rules. pronouncing only the article already triggers the execution of one of the variants of the intent
You need to run more tests to verify this assumption.