Yes, he really does sound like the Joker… 1 of 'em, anyways…
I think the argument should be about trustworthy, not quantity, of code.
iow, it doesn’t get committed until some human has read & understood it, & has signed-off on it.
Nobody signs-off on it?
Then don’t trust it.
LLM’s can outcode us humans…
but it can also code mistakes at something like 1-million times our speed, too!
Responsibility & accountability still are required, for any trustworthy infrastructure or civilization.
& since LLM’s are being programmed on trash, then they aren’t going to be entirely-trustworthy.
Were I to want an LLM to do some engineering, then I’d program it ONLY on good-code/good-data, until its “bones” were set, and then show it what to not do, to reinforce those instincts…
not just throw it at most used code, & deem that to be good-enough.
Yes, he really does sound like the Joker… 1 of 'em, anyways…
I think the argument should be about trustworthy, not quantity, of code.
iow, it doesn’t get committed until some human has read & understood it, & has signed-off on it.
Nobody signs-off on it?
Then don’t trust it.
LLM’s can outcode us humans…
but it can also code mistakes at something like 1-million times our speed, too!
Responsibility & accountability still are required, for any trustworthy infrastructure or civilization.
& since LLM’s are being programmed on trash, then they aren’t going to be entirely-trustworthy.
Were I to want an LLM to do some engineering, then I’d program it ONLY on good-code/good-data, until its “bones” were set, and then show it what to not do, to reinforce those instincts…
not just throw it at most used code, & deem that to be good-enough.
_ /\ _