A New Trick Uses AI to Jailbreak AI Models—Including GPT-4

-
Adversarial algorithms can systematically probe large language models like OpenAI’s GPT-4 for weaknesses that can make them misbehave.

from Security Latest https://ift.tt/PuMJHpQ

Comments

Post a Comment

Popular posts from this blog

Tesla backtracks on hiring someone to fight Elon Musk’s Twitter critics

-
Image
Reports published yesterday suggested that EV maker Tesla was on the hunt to hire someone specifically to fight against social media messages directed at the company’s CEO Elon Musk. However, it seems the company has already gone back on the idea. It hasn’t even been 24 hours yet, and the job description for a customer service post at Tesla has been changed to now omit any mention of them being required to fight messages leveled at Musk. [Read: How Netflix shapes mainstream culture, explained by data] As spotted by Jay Boller on Twitter, the original job posting required an “Energy… This story continues at The Next Web Or just read more coverage about: Tesla , Twitter from The Next Web https://ift.tt/3bUYUlR
Read more

The week’s best Android games for a Friday night in

-
Image
Welcome to our weekly column, in which YouTuber NimbleThor brings you a short list of the best new mobile games for your Android device. Whether you’re into casual time wasters or hardcore shooters, you’ll find something in here to exercise your thumbs with. So the weekend’s almost here, and you don’t have plans. Hey, we’re not judging – sometimes you just need a little downtime. Whether you’ve got a babysitting gig to attend to, or intend to simply coast through the weekend in a Homecoat, we’ve got a bunch of great new Android games without time-limiting energy systems for you to curl up with.… This story continues at The Next Web Or just read more coverage about: Android from The Next Web https://ift.tt/2PbO1wL
Read more