atomic latch
Why can't I ask the assistant longer questions? How can I change this so that I can ask him extended questions, etc.? Currently, I can ask him a question for about 5 seconds. And how can I teach him not to ask if he can help me with anything else, etc.?
stable talon
If you're pausing during your question and it stops listening, change this to Relaxed:
In the ESPHome device
atomic latch
And there is still the issue of longer answers, and teaching him not to say, for example, can I help you with something else?
mighty quarry
And there is still the issue of longer answers, and teaching him not to say, for...
Well this is dumb LLM. You have to teach it things.
atomic latch
HOW? When I told him not to say it, he still does it.
atomic latch
Well this is dumb LLM. You have to teach it things.
And why can it be that I can only ask him questions / commands for max 5 seconds. Is it chatgpt 3.5's fault?
stable talon
Did you change the thing I suggested?
atomic latch
Did you change the thing I suggested?
Yes
stable talon
That will wait for 2s of silence, with a timeout of 15s
atomic latch
Wait, i try
merry lodge
You're not going to replicate ChatGPT's Voice Mode using HA Assist if that's the expectation. When you say you "told it not to" do you mean in the prompt?
atomic latch
I mean, when I called him out, I told him never to do that. I will test this aggressive mode, because I have a problem with it at the moment.
stable talon
You can't just tell it not to do things in the future
It's like a cat
You can incorporate stuff like that in the initial prompt context
atomic latch
You can incorporate stuff like that in the initial prompt context
And how to do it? So that he doesn't do it
mighty quarry
And how to do it? So that he doesn't do it
Include it into the system info in configuration.
atomic latch
Include it into the system info in configuration.
On Openai configuration?
Include it into the system info in configuration.
mighty quarry
Yes.
merry lodge
And prompting it to be called "Homie" would not be useful here. This will not replace the wake word, and it would never really come up in conversation. But defining how it should respond, and how it should not respond, would be appropriate there.
atomic latch
Yes.
Well, I've done it before. And I gave him instructions to never do it again. But he still does it anyway.
merry lodge
Prompt it better. Ask chatgpt to improve the prompt you want. Explain what you are using it for. Experiment. There are no hard rules for any of this.
mighty quarry
Yes, try to explain your needs with free speech, like you would to 5-year old kid.
