Janelle Shane asked ChatGPT to play tic-tac-toe, and it consistently lost, while "explaining" why it did things. Given the board

O _ _
X X _
_ _ _

it ignored player 1's obvious winning move and put an O in a different corner.

The algorithm did live up to its reputation by offering pompous, long-winded, "explanations" for its decisions.

I'm told, by people who actually know the medical field, that LLMs are doing well at medical diagnosis. This latest result gives me doubts, moreso than ChatGPT inventing legal citations. At least tic-tac-toe is low-stakes.
Last night, the National Weather Service's Webpage included a discussion of Hurricane Fabian that said, in effect (I don't remember the exact wording) "it's not clear why the forecasting models say X, but we'll accept it."

Our software is starting to know things that the human experts don't. That's a step toward AI, and a useful one: machines that can do something valuable that humans can't do ourselves, namely predict how hurricanes will behave.
Last night, the National Weather Service's Webpage included a discussion of Hurricane Fabian that said, in effect (I don't remember the exact wording) "it's not clear why the forecasting models say X, but we'll accept it."

Our software is starting to know things that the human experts don't. That's a step toward AI, and a useful one: machines that can do something valuable that humans can't do ourselves, namely predict how hurricanes will behave.
.

About Me

redbird: closeup of me drinking tea, in a friend's kitchen (Default)
Redbird

Most-used tags

Syndicate

RSS Atom
Powered by Dreamwidth Studios

Style credit

Expand cut tags

No cut tags