News for hackers. Stuff that matters.

A comment on Training language models to be warm and empathetic makes them less reliable. Parent comment

Re:Training language models to be warm and empathetic makes ... by astrange on Tuesday August 12, 2025 @05:45PM (#44882157)
> Point out every unstated assumption

What, all of them? That's a difficult problem.

https://en.wikipedia.org/wiki/Implicature

> every logical fallacy

They killed Socrates for that, you know.

[ Parent | on HN ]
Re:Training language models to be warm and empathetic makes ... by IncreasePosts on Tuesday August 12, 2025 @06:22PM (#44882471)
I wonder if they would have killed Socrates if he proposed a "suitable" punishment for his crimes, as was tradition. He proposed either being given free food and housing as punishment, or fined a trifle of silver.
[ Parent | on HN ]
Re:Training language models to be warm and empathetic makes ... by potato3732842 on Wednesday August 13, 2025 @08:00AM (#44887384)
>He proposed either being given free food and housing as punishment, or fined a trifle of silver.

Contempt of state process is implicitly a crime just about everywhere no matter where or when in history you look so it's unsurprising they killed him for it. He knew what he was doing when he doubled down, probably.

[ Parent | on HN ]