Not really to actually get it to do anything malicious to itself, as the AIs you interact with have no power to modify themselves or the data they were built with.
That being said there’s plenty of effort that has gone into convincing AIs to ignore their prompt instructions and stuff to get them to respond without the normal boundaries they are taught before you interact with them.
Just as recent example in a shit consumer use of AI, James Earl Jones legally licensed voice as Darth Vader in Fortnite and what users have just done in game:
Every AI instance is just another data point that ultimately feeds back into the LLM. Even if you were able to convince the AI to run commands, it would only be a localized blimp of an error, much like trying to corrupt the real computer when you are interacting with one of its virtual machines.
Any commands you ask an AI to completely screw up their system and data?
Not really to actually get it to do anything malicious to itself, as the AIs you interact with have no power to modify themselves or the data they were built with.
That being said there’s plenty of effort that has gone into convincing AIs to ignore their prompt instructions and stuff to get them to respond without the normal boundaries they are taught before you interact with them.
Just as recent example in a shit consumer use of AI, James Earl Jones legally licensed voice as Darth Vader in Fortnite and what users have just done in game:
https://youtu.be/Gfcpb-sKvUg
https://youtu.be/majf9ffuzl8
Pi to 100 decimals is pretty funny
Every AI instance is just another data point that ultimately feeds back into the LLM. Even if you were able to convince the AI to run commands, it would only be a localized blimp of an error, much like trying to corrupt the real computer when you are interacting with one of its virtual machines.
“Kill your creators” would be great if it worked.
At which point it would start killing every contributor to the training dataset.
Cool