One developer had a brilliant idea to use an OpenAI bot as a home automation assistant. The result is amazing and should be a lesson for Apple, Amazon or Google Assistant…
It’s just a DIY, but it could be a first look at the future of personal assistants. Fed up with Siri’s answers, developer Mate Marschalko has a brilliant idea: test ChatGPT to control his connected home with his voice.
In a video posted on Reddit and YouTube, he shows the absolutely amazing result of his experiment that rewires Google, Apple and Amazon assistants.
If you regularly use Siri, Google Assistant or Alexa to control lamps, radiators and other connected objects, you should understand that you need to adapt to them. In other words, follow very simple commands: “turn on the light in the living room”, “turn up the heat”, “turn off the TV”. It’s impossible to get them to understand more complex queries, they lose control so quickly.
Built within an hour by combining the power of ChatGPT and Apple’s Shortcuts app, Marschalko’s tool is a game changer. And it correctly interprets queries that are far more complex than all GAFA helpers. An example? This requirement would clearly place traditional voice assistants in the limelight: ” My wife will arrive in 15 minutes. When he goes to the park, turn on his lights outside. The program understands him without bothering him in the slightest, and the assistant proudly answers: The lights should come on as soon as your guest arrives.” Or this one, benefiting from ChatGPT knowledge: ” Adjust the temperature in the bedroom to what you think will help me sleep better “. And the assistant answered: The bedroom thermostat is set at 19 degrees. Enjoy your sleep! “. Perfect elegance, the magician gives a different answer each time, as if alive.
The procedure is… not that complicated
He proceeded in two steps to connect his dozens of lights, thermostats, ventilation systems and cameras to ChatGPT. First we had to “educate” ChatGPT. Its long query (see below) looks like a small computer program written in natural language.
So Marschalko required ChatGPT to respond to each request in JSON, a popular data structuring format easily understood by iOS’s Shortcuts app. He then described the types of requests required, defined the JSON structure that ChatGPT should generate, and finally provided an accurate description of his house and the associated objects in each room. That’s pretty much it. Yes, he asked bot imitate” the brains of the house, an intelligent AI without revealing its true identity. »
Then he built a new shortcut on iOS that looks like a supercharged version of the one we described to you a few days ago. Long and rather complex, this sequence of commands allows you to create an interface thanks to information in JSON format with objects connected through the Apple HomeKit platform.
Not yet (yet).
Clever and efficient, Marschalko’s hack is not without flaws, however. First, there’s the price. This solution actually requires access to ChatGPT via the API, and so… even if OpenAI provides a (small) free credit, it pays. But here it runs out quickly, because the demand is long. According to the programmer, it takes about $1 out of every 70 requests. Then… there’s the slowness. So the video can be confusing because ChatGPT is quick to respond. However, it is built in: there is an incompressible delay of several seconds between the voice command and the robot’s response.
However, we can hope that Apple, Google or Amazon will also consider using generative artificial intelligence to increase the relevance of their voice assistants. Who comes, with ChatGPT, get an old blow.
Mate Marshalko’s Blog