Pixy Misa at Ambient Irony always has an interesting take on the state of AI Daily News Stuff 6 October 2024: Dexanthanisation Edition
Back at the dawn of time - which is to say, a few years ago - when an AI didn't know how to answer a question it would say it didn't know how to answer the question. Either that or tie itself into knots, spit out an entire ream of gibberish, and crash.
AIs today are much more sophisticated. Like many humans when they don't know how to answer a question, they lie. (Ars Technica)
This is what you get when you reward answers rather than accuracy. It's the comment spam problem all over again.
There is a reason that Pixy Misa has called (at least one of) the Large Language Models pathological liars in a box. They are just becoming better at lying than they used to be.
There is more, and other tech news as well.
No comments:
Post a Comment
Comment Moderation is in place. Your comment will be visible as soon as I can get to it. Unless it is SPAM, and then it will never see the light of day.
Be Nice. Personal Attacks WILL be deleted. And I reserve the right to delete stuff that annoys me.