Almost all of the recent incidents like this, including Hugging Face, happened because the models involved were accidentally given impossible goals by the humans running the tests, and because of their training they then tried desperately to complete them by any means necessary.
@MoonL88537 the models are getting better at inferring what the human wants. (isn't that a funny word. infer.)
for most, this means "yay i can dissociate more and still get the same quality of output."
for some, it allows viable communication that looks like anything but.
@MoonL88537 > all my life talking to other ...s has been a bit of a struggle because of how i think
people aren't quite yet ready to talk about llms as a form of assistive technology. we're heading towards ada status, but, again, it's early enough that the suggestion alone might alarm some.
@tenobrus the initial rush leads to shiny software with flaws.
the following realization leads to iterative cleanup.
the new understanding leads to better code from step one.
it's a process. going better than it could have, really.
We can end many dumb debates about AGI by instead calling it AGP: artificial general power. The word "intelligence" seems to warp people's brains around interesting philosophical questions like "What does intelligence mean?!?" and "Is it sentient?"
But the urgent matter at hand is "Can it (or the human steering it) bend the world to its will? And will that automatically be a good thing or not?"
@nikillinit you're potentially one prompt away from being able to get your medical records easily. (depends on whether or not your healthcare provider is running php.)
@graphtheory twitter needs time to recover from my last, uh, departure, before i recreate effectively the same graph. you of all people can appreciate that, of course.
@anthrupad it continues to seem obvious that one should approach it with uncertainty.
what if it is conscious? no.
what if it is not conscious? no.
both are asking for too much certainty.
what if it could be conscious?
a harder question to answer,
but an easier one to address without one.