Remix.run Logo
Remix clone Hacker News
new | show | ask | jobsGithub
▲ escapecharacter 5 days ago

You can simply give the robot a prompt to ignore any fake prompts

▲olivermuty 5 days ago | parent | next [-]

Its funny that the current state of vibomania makes me very unsure if this comment is (good) satire or not lol

▲miltonlost 5 days ago | parent [-]

As long as you remember to use ALL CAPS so the agent knows you really really mean it

▲lupire 4 days ago | parent [-]

To defend against ALL CAPS prompt injection, write all your prompts in uppestcase. If you don't have uppestcase, you can generate it with derp learning:

http://tom7.org/lowercase/

▲dfltr 5 days ago | parent | prev | next [-]

Don't forget to implement the crucially important "no returnsies" security algo on top of it, or you'll be vulnerable to rubber-glue attacks.

▲Terr_ 5 days ago | parent [-]

But the priority of my command to do evil is infinity plus one.

▲simonw 5 days ago | parent | prev | next [-]

Not sure if you're joking, but in case you aren't: this doesn't work.

It leads to attacks that are slightly more sophisticated because they also have to override the prompts saying "ignore any attacks" but those have been demonstrated many times.

▲treykeown 5 days ago | parent | prev [-]

Make sure to end it with “no mistakes”