How One Developer Combined the Mind of ChatGPT with the Voice of Apple’s Siri
A developer augmented his Apple Siri smart home with the power of GPT-3, melding the conversational abilities of ChatGPT with the voice interface and device controls of Apple HomeKit. Mate Marschalko’s AI Frankenstein performs remarkably well, as seen in the video above, and might hint at how voice assistants will evolve to employ the LLM-derived conversational AI.
Marschalko set up the demonstration by linking GPT-3 to Siri through Apple’s Shortcuts. He explained to GPT-3 in plain, though detailed, language the kinds of requests he might issue to Siri and how it should respond in the JSON computer language compatible with a smart home AI. He described some of the information to include in the response and a list of details about the house, its smart devices, and even the home’s location.
The initial test had Marschalko tell Siri, “Just noticed that I’m recording this video in the dark in the office, can you do something about that?” The voice assistant passed his words on to GPT-3, which created a machine-readable code for turning on the lights in the office and sent it back through Siri to Shortcuts, activating the light. The video edits out a few seconds of response time for each request, but GPT-3 boosted Siri’s ability to respond to far more indirect and casual speech than usual.
“Honestly, when I first saw this response I couldn’t believe my eyes and how exceptionally well it worked! The request was not a simple “Switch the lights on the office.” It was phrased in a very twisted and elaborate way. Something that would immediately throw off Siri, Alexa or Google Home,” Marschalko wrote in a Medium post about the demo.”Another interesting example was how it was able to make a decision instead of me on what temperature I should set the bedroom to [by saying] “I’m going to trust you with this one! Set the bedroom to a temperature you think would help me sleep better.” And it set the bedroom to a comfortable 19 celsius based on its knowledge!”
Experiments with GPT-3, ChatGPT, and other generative AI models have produced a lot of entertaining and occasionally useful projects, like allowing an artist to converse with her younger self or bringing Santa Claus to life. Still, Marschalko single-handedly raised Siri’s conversational intelligence to an astonishing level while simultaneously showcasing GPT-3’s ability to communicate with both people and machines without any coding knowledge necessary on the human end. If you’d like to try it out, Marschalko shared the Okay Smart Home Siri Shortcut he made along with the prompts he used to set up the demonstration. If you test it out, drop me a line and share how it went.