18 comments

[ 2.8 ms ] story [ 48.3 ms ] thread
Some of those "answers" have to be scripted
I'm not sure it's necessarily scripted -- maybe it was just seeded very carefully, so for each keyword it doesn't have much to choose from.

But clearly it's saying a lot of things without understanding. Probably nothing special here.

Yeah, but he could ask the questions in any order!
unless we can interact with it (Silvia) ourselves... it's really hard to believe

This makes it a little more believable: http://cognitivecode.com/products.html#studio

> unless we can interact with it (Silvia) ourselves... it's really hard to believe

Agree.

> This makes it a little more believable: http://cognitivecode.com/products.html#studio

Not really. You can't really download or get it until some future release date, so who knows how good this thing is.

Didn't watch till the end, but in the first few minutes it did not show any learning, or did it? That would have been interesting, like SILVIA doesn't know an answer to something, guy explains, SILVIA understands and reiterates in her own words.

I am not holding my breath on this one. Especially, why did they bother to create the face and voice recognition? Seems to me creating good conversations is hard enough, I don't see anybody making a breakthrough in that having spare time to create a pretty talking face thingy.

That was altogether pretty lame. Having done some NLP and chatterbot development in the past, I don't see anything here out of the ordinary. In fact, it seems as though they're just trying to impress folks who don't know any better.

People like them (Cognitive Code) give AI research a bad name. As always, put up or shutup.

His questions covered only very narrow territory. But she was able to connect different parts of the conversation quite well. If you call that lame you're going to run out of words to describe 90 % of the software that's being created (including much of what I've done myself to be honest)

Sturgeon's law: 90% of everything is crap.

This applies especially to attempts at NLP.

This shows nothing about A.I. Only some good speech recognition software.
I want to believe, but an prepared for disappointment.
agree, pretty lame that they dont show a believable, casual conversation with someone not related to the project

still, this bot seems many times more articulated than me :P

What they have is impressive at least as an implementation of AI window dressing. It did a decent job with context assuming this was not overtly scripted as they claim.

Most notably, though, if sylvia was an astounding breakthrough in the sense they claim (strong AI) they would have a sylvia chat-window on the website. The absence of a chat window means they realize if you were to poke at it you would be let-down. If anyone finds a sylvia instance to chat with, please post it.

http://cognitivecode.com/

extraordinary claims require extraordinary evidence.
You know that if you ever got to use that when no one was looking over your shoulder that you'd immediately start saying the most obscene things you could think of. Everyone does.
That would be a better turing test. If you talk dirty to it, and then you feel bad about it after the fact (because of how it responded) - then it has passed .

Some may say "aroused" by it but the thought is too disturbing.

"Is it afternoon already? I'm afraid I don't have a good sense of time."

You're telling me that "she" didn't know what time it was?