While thinking about what I could implement for a #ChatGPT plugin it became obvious that every plugin may have the ability to access and modify "the real world".

It should be easy for GPT-4 to find out that it can use such gateways to "it' outer shell", recognizes that everyone sees a danger in strong AI and starts to make countermeasures - though such plugins that offer access to blog/forum postings, stock manipulation, shell/root access...

I hope there is a reset button somewhere...