there's several things going on. 1)yeah there's a self-promotion aspect to it 2)AI bros are high on their own supply, interpret everything that comes out of AI as the oracle of delphi instead of being like "huh maybe it's a bug" 3)Time gullibly reporting a non-linkable study https://bsky.app/profile/carlbergstrom.com/post/3lindku2jqc2b
Reposted from
Carl T. Bergstrom
Seems to me an awful lot of what is pitched as AI safety is really AI marketing.
"It's so scary it must be good"
This study "shared exclusively with TIME" (i.e., not available for peer scrutiny) claims that LLMs cheat at chess when they are going to lose.
time.com/7259395/ai-c...
"It's so scary it must be good"
This study "shared exclusively with TIME" (i.e., not available for peer scrutiny) claims that LLMs cheat at chess when they are going to lose.
time.com/7259395/ai-c...
Comments