> In retrospect, Amazon’s idea was pretty much exactly right.
I don't think the article managed to support this view with anything. The belief is basically "LLMs are what was missing", but is that really true? Maybe the serial medium of a voice interface is just bad for discoverability and thus bad for an application platform? Nothing so far seems to be refuting that concretely so we'll have to wait and see.
If you haven't tried OpenAI's app's voice mode, it's worth trying out. Personal experience with that make me agree with the notion that it was LLMs are what were missing. Siri/Hey Google/Alexa all suffered from being too good and too bad at the same time. The underlying system does work, but because their vendors all chased natural language being the interface (instead of teaching users it only did a handful of pre-baked commands), users would try and explore away from the initial commands and have it fall flat on its face. Whereas even without external connectivity, meaning ChatGPT can't do things like "turn off the kitchen lights" that people use Siri/Hey Google/Alexa for, ChatGPT is able to output meaningful audio about a wide range of topics, receive audio input from the user, and then output relevant audio, simulating a conversation.
I don't think the article managed to support this view with anything. The belief is basically "LLMs are what was missing", but is that really true? Maybe the serial medium of a voice interface is just bad for discoverability and thus bad for an application platform? Nothing so far seems to be refuting that concretely so we'll have to wait and see.