Frustrated with Siri’s inability to understand your queries? There is a way to get Apple’s voice assistant to work with it GPT-3 (OpenAI’s deep learning language model used in the AI chatbot chat) to become more conversational and smarter.
Developer Mate Marshalko has used Siri’s voice, Apple’s Shortcuts app, GPT-3 intelligence, and a bit of hacking to create a surprisingly smart home assistant that can respond to rather tricky queries.
In a video posted on reddit (via the edge), we can see Marshalko issuing increasingly complex (and not direct) orders to his assistant. In one example, Marshallko says “Yesterday it took me a while to fall asleep. Do you have any suggestions to help me sleep better?” So, the assistant responds by suggesting a warm bath, reading a book, and avoiding caffeine. In another example, Marshalko says, “Notice I’m recording this video in the dark, in the office. Can you do something about that,” and the assistant quickly turns on the light.
Compared to Siri, Alexa, and Google Assistant, this home assistant seems to understand complex queries better, but also responds in a manner much like talking to a real person.
Marshallko says building this smart home assistant was a fairly simple task. “I was able to achieve all of this by simply asking GPT-3 in my post pretending to be my home assistant, listing items in my home, and a few other details about time and location, and then asking her to respond in a structured, structured data format (JSON) which I could then use To trigger HomeKit control messages in a series of if…other statements in a single Siri Shortcut.
Yes, it’s a bit of work (detailed instructions here), and it does require some basic programming knowledge, but almost anyone can do it in less than an hour (according to Marshalko).
While this demo seemingly puts Apple’s Siri to shame, things aren’t quite so simple. API requests to GPT-3 cost about $0.014 per request, so unlike Siri, this smart home assistant isn’t free to use. And Siri, like other AI assistants, is designed not to be rude, obscene, or do anything that might be considered harmful; A language paradigm like GPT-3 has some safeguards, but it can certainly respond in unexpected ways. Finally, without testing the bot ourselves, it’s hard to say how often it goes wrong (in Marschalko’s video – modified – it responds perfectly every time).
Still, it’s an incredibly cool experience, which will undoubtedly lead to more homemade Siri clones. We’re excited (and a little scared) to see what can be done with this collection.