logoalt Hacker News

Barrin92today at 5:57 PM1 replyview on HN

>We can have actual conversations with our computers

conversational interfaces we've had for decades. In fact this goes so far back that people thought ELIZA was sentient (https://en.wikipedia.org/wiki/ELIZA_effect), and not just laypersons but even the people working on it.

The actual lesson from this stuff is that linguistic interfaces fry people's brains, you could convince nerds that a brick was intelligent if you hooked it up to the voice of Scarlett Johansson. The perception that these systems are in any meaningful way intelligent, when they start getting stuck in a doom loop of fixing the same two bugs by reintroducing them in circles or give entire reviews for a music album that doesn't actually exist, is entirely in the head of the user.


Replies

queenkjuultoday at 9:15 PM

What's nuts to me is how sycophantic the models are even when I'm just generating code. "What a fantastic project!" "Such an elegant solution!" When all I've done is described the problem and pasted a debug trace or whatever.

It's annoying in my context but it's no wonder people jumping into open ended "conversations" with these things and up in dangerous feedback loops