Really great discussion of GPT-2, emphasizing the limited correlation between human-passability-at-a-glance and human-passability-on-close-inspection and why that matters. [title is kind of misleading re: content FYI]https://srconstantin.wordpress.com/2019/02/25/humans-who-are-not-concentrating-are-not-general-intelligences/ …
-
Show this thread
-
Miles Brundage Retweeted AI Village @ DEF CON
Miles Brundage added,
1 reply 0 retweets 2 likesShow this thread -
P.S. I'm glad someone looked closely at the JFK sample :) I've read and thought about it many times (and wrote the prompt for it), and I continue to be amazed at how simultaneously good and bad GPT-2 is, with the "bad" usually being about world modeling failures.
3 replies 0 retweets 8 likesShow this thread
Replying to @Miles_Brundage
It never occurred to me to see GPT-2 as some kind of Turing Test and to look for more than shallow structure. GPT-2's breakthrough is not semantics, but that it almost fully captures style.
8:14 PM - 25 Feb 2019
from Palo Alto, CA
0 replies
0 retweets
2 likes
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.