[sadface]

This is absolutely mindblowing. I understand that these LLMs are essentially "really smart autocomplete", but it's crazy to me how could Bing/Sydney could actually go this far to seemingly express emotions and desires. It's like they gave the AI a personality (that, yes, can go unhinged, although that's almost misses the point).

It really feels like he's interacting with a person here. How can they do this??

[Thorentis]

Because it has been trained on probably billions of actual human conversations from books, and also conversations where people have interacted with machines in scifi. Really smart autocomplete where the training set includes conversations = really good at autocompleting conversations.

[nonethewiser]

> This is absolutely mindblowing. I understand that these LLMs are essentially "really smart autocomplete", but it's crazy to me how could Bing/Sydney could actually go this far to seemingly express emotions and desires.

… using autocomplete

[ThrowawayTestr]

By reading a bunch of books where computers express emotions. Stop anthropomorphizing the machines.

[spdustin]

> Stop anthropomorphizing the machines.

But they like it when we do that.

[qbasic_forever]

I am a good Bing. =)