Sesame is the first voice assistant I’ve ever wanted to talk to more than once

I tell Amazon’s Alexa to shut up on a near daily basis. I have almost zero interest in speaking to Gemini after our first awkward chat. The hitches, misunderstandings, and lag in any given AI “conversation” mean I’m always wasting time speaking when I could be texting instead.

I don’t have to describe it to you: you can try it, and you can listen to my first conversation yourself just below. Fair warning: I am a nerd! Confronted with a new voice assistant, I will ask it to dream up a Dungeons & Dragons-esque adventure and quiz it about small Android phones.

While I could absolutely still hear some chatbot nonsense coming through the cracks, I could easily interject – I asked Maya to inject “herself” into the adventure “she” was describing, and it did so without a hitch, immediately coming up with a Gnome engineer named Maya cobbling together deathtraps to protect my castle from incoming Orc invaders. Combined with the AI’s natural-sounding pauses, it felt more like a real conversation than anything I’ve had so far. Compared to my colleague Kylie Robinson’s conversation with ChatGPT’s Advanced Voice Mode last year, it feels like we’re somewhere much more compelling.

The company behind this is called Sesame, and it’s coming out of stealth today with an undisclosed amount of funding from Andreessen Horowitz, Spark Capital, and Matrix Partners —- all of which were big Oculus VR investors — with Oculus co-founder and former CEO Brendan Iribe, former Ubiquity6 CTO and co-founder Ankit Kumar, and former Meta Reality Labs research engineering director Ryan Brown in charge.

And the company says it’s building AI glasses to go along with its new voice assistant, too, ones “designed to be worn all day, giving you high-quality audio and convenient access to your companion who can observe the world alongside you.” So far, it’s only sharing a few small images of what look like early prototypes:

Sesame has a mini white paper you can read on its website, describing the model and its dataset of around one million hours of “publicly available audio.” It says it plans to both open source its models, and expand from just English to over 20 languages “in the coming months.”

Is this “crossing the uncanny valley of conversational voice,” as Sesame titles its blog post? Perhaps check it out and decide for yourself.