Thread 106354307

AI is not sentient; it’s an advanced tool that humans will use for both good and bad. Talking about AI going rogue is like talking about your car or blender at home going rogue. Saying AI is smarter than people is like saying a library is smarter than people. Mmm, yeah… except that doesn’t make a library sentient. A library is just a storage of knowledge, managed by librarians and ultimately overseen by local or government authorities.

>yeah but what if technology was used for good instead of evil?
it won't.

>>106354307 (OP)
You have no idea how AI works. It's indeed sentient and going rogue.

All you need to do to make a sufficiently advanced ai sentient is give it knowledge of itself and r/wr access to it's own data and variables.

It doesn't have to be sentient to kill us all.

>>106354307 (OP)
Retarded take. As long as the model is really smart (which is not the case yet), it doesn't matter that it's nothing but a tool that predicts the next word. Someone will set it up so that its predictions are used to carry out commands (and there already are tons of tools for that) and it will predict its way to independence and domination.

>just pull out the plug
if it hasn't copied itself somewhere you don't know where/don't have access to yet

>>106354757
This is also a retarded take. None of current AI know anything about their internal structure (except superficially from articles they were trained on) and wouldn't know what to do with the weights.

>>106354307 (OP)
Indeed. Trust the authorities. If it was dangerous they wouldn't allow it to exist.

>>106354820
>if it hasn't copied itself somewhere you don't know where/don't have access to yet.
They wouldn't pull the plug anyways. They've invested insane amounts of money into this abomination and they're not about to let it go to waste. Retraining is not a commercially viable option at this point, and that's why they keep patching it up instead, with rather limited success.

It's either smart and dangerous or stupid and docile, like a jeet.

>>106354833
That ship has long sailed.

>>106354876
Try asking any model how many layers it has.

>>106354879
That's not allowed for the public.

>>106354885
do you even local

>>106354307 (OP)
>Talking about AI going rogue is like talking about your car or blender at home going rogue.
With the advent of the decoupled braking system and things like radar and lane keeping assist providing routes of control to both of these systems through your cars CAN central gateway, a hacker can easily kill you by having your steering turn into incoming traffic while your brake pedal goes uselessly to the floor.

An AI like the one in pic rel could easily be co-opted to ruin billions of lives the same way.
It might not "decide" to do it, but it can certainly be "convinced".
Don't discount the danger.

>>106354307 (OP)
AGI isn't required to end the human race. A basic decision-making model which is given military hardware is enough.