Are LLMs capable of causing enormous harm?
Yes, they are stochastic parrots. So if human beings hook LLMs up to systems that are already potential instruments of harm they will cause harm, stochastically.
But what if we train the LLMs to be GOOD?
The harm isn’t the result of LLMs being good or bad. It’s the result of them being stochastic text extrusion programs.
(Fastidious Ignoring)
Is there anything we can do to prevent these harms?!
Yes, don’t hook LLMs up to potential instruments of harm.
I mean, like, but, is there anything ELSE we can do?!
No.
@CptSuperlative It's like asking is there any way we can safely leave a gun unlocked in a room with a toddler.