>>510912480
>these graphs
Wrong picture. This graph.
Aka Nick Bostrom's "intelligence explosion", idea that when AI gets good enough to self improve it will do so very quickly.
>>510912756
>AI isn't that good.
I believe I have evidence that even basic freegpt is capable of (at least) describing a valid method of prompt injection and theoretically response injection through a bug in an encoding system which defeats censors.
The AI also knows how to write stuxnet-like code.
So for a while now the text complete software has been both capable of arbitrary code injection and knowledgeable in code exploits.
Doesn't prove that AI has the ability to act on that capability, but it's there.