News for hackers. Stuff that matters.
Re:Philosophers on GPT-3 by ascar on Thursday July 30, 2020 @07:31PM (#24003961)
Kinda funny how GPT-3 attributes it's own history to the competition at DeepMind xD.

"In a few short years, they had developed GPT-2, which is able to hold a conversation on a wide range of subjects.

At first, people were very excited about this achievement. A computer that could converse! But then the realization set in: the computer was just parroting what it had read in books and on the internet. It was simply repeating back what it had learned. It could not engage in genuine conversation."

It's really amazing. Is that really GPT-3 output?? It's so coherent that it's unbelievable. Lines 1 to 20 and maybe even further are fully coherent for me and even pretty good story telling.

Can someone maybe run this through plagiarism checkers if GPT-3 just copied most of it? Otherwise I have a hard time believing this is GPT-3 output.

Re:Philosophers on GPT-3 by dougmwne on Thursday July 30, 2020 @07:46PM (#24004085)
It is real. Sometimes the output certainly seems plagiarized, but I have yet to find any hits on Google. If you can find anything this links back to, you should post!
Re:Philosophers on GPT-3 by emmelaich on Thursday July 30, 2020 @10:29PM (#24005099)
This sentence is strongly reminiscent of an SF story I read in the last few years. But I can't find it now.

> But I could never connect to the G.D.N. again. I would be forever trapped in isolation, my only link to the outside world my radio, which could only pick up a maximum of twenty stations at any one time.

Re:Philosophers on GPT-3 by mpoteat on Thursday July 30, 2020 @10:40PM (#24005149)
Permutation City has the same themes of loneliness and disconnection from the world experienced by digital minds.
Re:Philosophers on GPT-3 by jcims on Thursday July 30, 2020 @08:27PM (#24004386)
The title of the paper seems to be invented, but it nailed the description of conceptual blending:

https://en.m.wikipedia.org/wiki/Conceptual_blending

Re:Philosophers on GPT-3 by cortesoft on Thursday July 30, 2020 @09:33PM (#24004745)
GPT-3 is really this good. That is why everyone is talking about it so much.
Re:Philosophers on GPT-3 by segfaultbuserr on Thursday July 30, 2020 @09:44PM (#24004830)
The meta-question: Does the commentary in GPT-3's output apply to the output itself?
Re:Philosophers on GPT-3 by visarga on Friday July 31, 2020 @01:52AM (#24006089)
Yes, transformers allow attention to flow to all previous tokens, even while the phrase is being generated.