If it fails, reply that it screwed up and needs to fix it. I bet that would work.

I'm at work so I don't have the prompt handy, but I instructed chat GPT to output commands in the following format:

[deviceName:state]

So chatGPT might reply with:

> I turned on your bedroom light [bedroom light:on] and turned up the temperature [thermostat:72]

All you have to do is parse the messages for [:] and trigger the thing.

EDIT:

I told it to place all commands at the end, but it insists on inlining them. Easy enough to deal with.

Prompt. Thanks. Damn auto correct needs ChatGPT level intelligence.

I didn’t hook up voice because of that awkward part. If I could get my hands on a raspberry pi I might make my own listening device.

I crafted a prompt to get ChatGPT to act as a home automation assistant. I told it what devices we have in the house and their states. I told it how to end any statement with one or more specially formatted commands to manipulate the accessories in the house.

It was just a fun POC, but it immediately became clear how much better this could be over Alexa or Siri.

I was able to ask it to do several things at once. Or be vague about what I wanted. It got it.

https://youtu.be/DOWDNBu9DkU These suckers are super quiet. And at the altitude they stay at, you don’t hear them.

These particular ones are very quiet. Check out the funky blade design.