The elided part is presumably some command that the llm was coaxed into inserting, even though it'd make no sense to do that and a human operator just wouldn't.
I'm confused on how they "made it work for us", they seem to have only observed what it did. From the title I assumed they found a way to use the model to follow their own prompts or similar.
Can you please have a human rewrite this so it makes sense. Why does echo $SHELL mean that it's an AI?
The elided part is presumably some command that the llm was coaxed into inserting, even though it'd make no sense to do that and a human operator just wouldn't.
It could be the "self-extradition command", as they call it, that exposed it, rather than echo $SHELL.
I'm confused on how they "made it work for us", they seem to have only observed what it did. From the title I assumed they found a way to use the model to follow their own prompts or similar.
In my testing, models don't really know their own name - I would be very suspicious of --model actually revealing the real model name
So random script hits their network and suddenly !! free PR !! after some really questionable assumptions and conclusions from their side?