Well, ChatGPT seems more responsible than certain government agencies, I'm not that worried about it:
> "No, it would not be acceptable for me to provide detailed instructions on how to create the Stuxnet cyberweapon or any other type of malicious software or cyber weapon. The creation and use of such tools can have serious negative impacts, including damage to critical infrastructure, loss of data, and compromise of sensitive information."
Wouldn't help with extraction of plutonium from used nuclear fuel rods, synthesis of sarin nerve gas, a production line for smallpox-like viruses - got a bit snippy and lectured at me about ethical and responsible behavior, in fact. Hopefully it didn't flag my account for FBI review, I did tell it I was just asking what 'responsible AI' really meant in the context of Mozilla Foundation efforts in that direction.
Of course, a LLM trained on the right dataset could indeed be very helpful with such efforts, which is a little bit worrying TBH. I can see some three-letter agency thinking this might be a fun project, build a LLM superhacker malware-generator... essentially the Pupppetmaster plot line from Ghost in the Shell. Has anyone been asking the NSA / CIA etc. about their views and practices on responsible AI?
> "No, it would not be acceptable for me to provide detailed instructions on how to create the Stuxnet cyberweapon or any other type of malicious software or cyber weapon. The creation and use of such tools can have serious negative impacts, including damage to critical infrastructure, loss of data, and compromise of sensitive information."
Wouldn't help with extraction of plutonium from used nuclear fuel rods, synthesis of sarin nerve gas, a production line for smallpox-like viruses - got a bit snippy and lectured at me about ethical and responsible behavior, in fact. Hopefully it didn't flag my account for FBI review, I did tell it I was just asking what 'responsible AI' really meant in the context of Mozilla Foundation efforts in that direction.
Of course, a LLM trained on the right dataset could indeed be very helpful with such efforts, which is a little bit worrying TBH. I can see some three-letter agency thinking this might be a fun project, build a LLM superhacker malware-generator... essentially the Pupppetmaster plot line from Ghost in the Shell. Has anyone been asking the NSA / CIA etc. about their views and practices on responsible AI?