Crossing the uncanny valley of conversational voice | Hacker News Reader

Crossing the uncanny valley of conversational voice

sesame.com

82 pointsnelwr14 comments

The voice is really good, especially small details like sharp air intake before a response to indicate tone. I guess the conversation topics weren't amazing but that's not really the point.

The problem I have is that if you want to engage with something like this, you need to pretend it's a human. As they say, the uncanny valley was pretty much successfully crossed. And to be honest, I don't want to pretend I'm talking to a human for a whole bunch of reasons.

The technically aspects are really impressive, but I think "pretending to be a human" in this way is a pretty scary goal. The cognitive dissonance was too strong and it was hard for me to continue a conversation very long at all. What does it even mean to have a conversation without theory of mind?

People talk a lot about how transformative AI will be but rarely from the point of view it will have on society when interactions like this are 10x more realistic and ubiquitous.

I imagine an agent with this kind of conversational capability but better. And then I imagine that coupled with a video model that presents a proper talking head that is pretty much indistinguishable from a real one

Then what if these personalities were highly configurable to your tastes, had a very long memory and were so cheap and available you could build an entire customised social circle from them.

This is a bizarre nightmare for us but will become the norm for future generations. As rigid and inflexible as it is, my kids even had a hard time understanding Alexa is not real at first, and they have come accustomed to this style of conversational interface from a young age, I think about that but in 30 years for their kids.

If I listen to robot voices for to long it gets unbearable. The fascinating part imho is that human voices apparently have other communication channels that bypass the conscious mind. Like regional accents contain a kind of collective mood but more granular. It must therefore be possible to encode a beneficial feeling into the robot voices. We could all be Irish :)

This isn't getting the attention it deserves. First AI I've had a real conversation with.

Agreed, I'm surprised that it hasn't gathered more buzz. It's the most impressive AI demo I've seen this year (in a year packed with impressive demos!)

This is the most realistic AI conversation I ever had. I legitimately felt I was just chatting with someone, getting to know them. The male was more convincing for me.

You can ask it to talk serious without any quips or follow on questions and it's just a more sycophantic chatbot with great interruption flow.

The realism is needed to unlock next-gen use-cases. Then you can trust it to be an easily adjustable medium on the fly while you do more things.

Doesn't work?

`Uncaught (in promise) TypeError: 'microphone' (value of 'name' member of PermissionDescriptor) is not a valid value for enumeration PermissionName.`

Just curious, how it is different from the tech used behind GPT advanced voice mode ? In my experience that also changes tone on the basis of the emotions and is able to maintain the flow pretty well.

Eh, in their demos maybe. In practice it's still quite limited.

AI waifu getting scary.

This voice chat is so realistic, I’m left speechless.

“Her” is here.

They will open source the weights under Apache2:

https://github.com/SesameAILabs/csm

This is incredible. I'm speechless.

And it'll call you out for the silence on the line...

Reply on news.ycombinator.com