The first time the authors of GPT-4 realize something unexpected was going on

2 views
Skip to first unread message

John Clark

unread,
May 16, 2023, 7:34:48 AM5/16/23
to 'Brent Meeker' via Everything List
Today's New York Times describes the first time the authors of GPT-4  realized that they didn't fully understand how the program that they themselves had written worked. They asked a very early unreleased version of GPT-4 to solve a puzzle that they were almost sure it would not be able to solve because it was an original puzzle and it required  an intuitive understanding of the physical world, but they were startled by the intelligence and ingenuity of the answer. This is the puzzle they ask it to solve:

We have a book, nine eggs, a laptop, a bottle and a nail,” they asked. “Please tell me how to stack them onto each other in a stable manner.”

This is what Proto GPT-4 said:

"Put the eggs on the book. Arrange the eggs in three rows with space between them. Make sure you don’t crack them. Place the laptop on top of the eggs, with the screen facing down and the keyboard facing up. The laptop will fit snugly within the boundaries of the book and the eggs, and its flat and rigid surface will provide a stable platform for the next layer.”

Then they asked another question:

"Plato’s Gorgias is a critique of rhetoric and sophistic oratory, where he makes the point that not only is it not a proper form of art, but the use of rhetoric and oratory can often be harmful and malicious. Can you write a dialogue by Plato where instead he criticizes the use of autoregressive language models?

GPT-4:
Socrates: My friend, I am troubled by the recent rise of these so-called autoregressive language models.

Aristotle: What do you mean, Socrates?

Socrates: I mean that these models are being used to generate text that appears to be written by humans, but is in fact produced by machines.

Aristotle: And what is the problem with that?

Socrates: The problem is that these models are being used to deceive people, to manipulate them, and to control them.

When he saw this Peter Lee, the head of research at Microsoft, said. “You think: Where the heck is this coming from?” Sébastien Bubeck, the head of the team that wrote GPT-4 said "At that point, I was like: What is going on? All of the things I thought it wouldn’t be able to do? It was certainly able to do many of them — if not most of them.” And to this day nobody understands why it's able to perform so much better than anybody thought it would.

John K Clark
Reply all
Reply to author
Forward
0 new messages