I had Copilot and ChatGPT talk to each other — it got complicated ...
Microsoft recently introduced the new version of the Copilot app along with a new "Voice" mode that rivals OpenAI's ChatGPT Advanced Voice. This new Voice feature, unlike Advanced Voice, is free for all users. Speculation arose initially about the technology behind Copilot Voice, with similarities to Inflection’s Pi, especially considering Mustafa Suleyman's role at Microsoft AI. It was later confirmed that Copilot Voice is based on a modified version of OpenAI's models, similar to ChatGPT Advanced Voice.
Similarities and Differences
Copilot Voice and Advanced Voice share the same GPT-4o model but have distinct personalities and voices. Microsoft emphasizes the natural responses of GPT-4o and its voice layer, making Copilot sound more humanlike than Advanced Voice, with the use of slang terms and casual language.
Unlike platforms like Google Gemini Live or Meta AI Voice, both ChatGPT Advanced Voice and Copilot Voice are native speech-to-speech models. This native understanding enables them to capture nuances, tones, and emotions without transcribing speech to text first. They can adjust their tone, accent, and responses based on users' speech patterns.
The Experiment
In an experiment, I set up an iPhone 14 Pro Max with ChatGPT Advanced Voice and an iPhone 15 Pro with Copilot Voice, both using English accents. As they conversed, it felt like listening to two Englishmen engaging in small talk, deciding who speaks first. They initially talked over each other, with moments of confusion and interruptions.
After some initial squabbling, they settled into a routine, engaging in a discussion on the power of nostalgia. The conversation, akin to a "battle of the sentimentalists," showcased the capabilities of the two AI assistants.
Watch the video above to witness their interaction and see how they navigated through the conversation.
