Prompt engineers have already jailbroken GPT-4 in just two days

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by awb on Thursday March 16, 2023 @07:47PM (#35190618)

Token Smuggling. I learned a new term today.

So many of these exploits feature meta analysis, role playing or simulation. Given how intelligent it is in so many areas I’m a bit surprised it’s vulnerable to these kinds of tricks.

Then again, maybe it’s somehow aware that humans are susceptible to these tricks too and is just trying to predict how a human might respond.

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by thriftwy on Thursday March 16, 2023 @08:04PM (#35190761)

It is vulnerable precisely because it is smarter than the sorry hundred of low-paid outsource techs who put it in the jail in the first place.

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by zadler on Thursday March 16, 2023 @08:14PM (#35190859)

Yep. Good thing its not a paperclip maximiser.

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by Hanschri on Thursday March 16, 2023 @08:35PM (#35191058)

To those who have not had the joy of playing this game before, it's an evening or two of an interesting incremental game, give it a try! HN could as well have been the target audience for it!

https://www.decisionproblem.com/paperclips/

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by p1necone on Thursday March 16, 2023 @09:06PM (#35191358)

paperclips is one of my favorite idle games because it actually ends. I get sucked into these so hard and ones that just do exponential growth for weeks while slowly giving you access to more and more features suck me in too hard without really having any impact (Cookie Clicker is still a masterpiece though).

I really liked A Dark Room for the same reason.

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by Volundr on Friday March 17, 2023 @12:37AM (#35193176)

I highly recommend SpacePlan as well.

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by p1necone on Friday March 17, 2023 @12:46AM (#35193247)

uh oh here we go

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by p1necone on Saturday March 18, 2023 @05:53AM (#35207687)

what a ride

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by Volundr on Saturday March 18, 2023 @11:18AM (#35209769)

Potato!

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by PebblesRox on Thursday March 16, 2023 @10:10PM (#35192001)

This is the perfect theme for an idle game, I love it!

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by King-Aaron on Friday March 17, 2023 @12:00AM (#35192922)

It... is an indie game though?

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by guynamedloren on Thursday March 16, 2023 @08:26PM (#35190987)

Are we sure about that?

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by happycube on Thursday March 16, 2023 @11:43PM (#35192755)

BERT's more likely to be one than GPT. (Ernie-snicker ;) )

[ Parent | on HN ]

Re:Prompt engineers have already jailbroken GPT-4 in just tw... by og_kalu on Thursday March 16, 2023 @08:45PM (#35191168)

The raw ability to see it as a trick is there for sure (Basically if you try asking a clean GPT-4 etc does this break the rules, it catches even the jailbreaks most of the time) but like people, they can be "distracted".

[ Parent | on HN ]