Follow

blazing hot AI take that techies are definitely not ready for, caps 

It is nearly impossible to train a neural network to do anything but recreate the status quo and you can certainly never teach it to have independent guiding values from the training data or anything resembling a social imagination. Using it outside of EXTREMELY NARROW, PURELY ADVISORY contexts where this can't really be much of a problem like assisting in the classification of medical imagery for predicting disease presence should be permanantly banned butlerian-jihad style.

:dont_at_me:

blazing hot AI take that techies are definitely not ready for, caps 

@thufie I boosted before I finished the first sentence.

This is much more thorough than my common refrain that "neural networks can't do anything humans can't already do" and gets more to the core of the issue

blazing hot AI take that techies are definitely not ready for, caps 

@juliana yeah and of course techbros love it when they have never had an original thought under the sun, just like neural networks.

blazing hot AI take that techies are definitely not ready for, caps 

@thufie fuck i'm dying because you're right

blazing hot AI take that techies are definitely not ready for, caps 

@juliana the only difference is their training data is reddit lmao

blazing hot AI take that techies are definitely not ready for, caps 

@thufie now i'm laughing loud enough to be embarrassed in this café

blazing hot AI take that techies are definitely not ready for, caps 

@thufie

This techie fully agrees.

blazing hot AI take that techies are definitely not ready for, caps 

@thufie

"Thou shalt not make a machine in the image of a human mind...

except for assisting legitimate medical and or ecological operations"

The battlecry of the Butlerian Jihad was never the same again

blazing hot AI take that techies are definitely not ready for, caps 

@thufie the only people who think about the potential of AI the way you're critiquing, in my experience, don't work with AI.

More often they are simply biz folks that sell the idea, which is mostly vapor

@dualhammers unfortunately this has been the opposite of my experience at university

@thufie at university are you working with techies actually working at A.I. companies or are you just encountering students who are studying A.I.?

@dualhammers the department head's specialty area is neural networks, even.

blazing hot AI take that techies are definitely not ready for, caps 

@thufie as a data scientist and ai researcher I must say: hard agree

blazing hot AI take that techies are definitely not ready for, caps 

@thufie what else would you call fancy regression?

blazing hot AI take that techies are definitely not ready for, caps 

@teh_dude people compare these things to BRAINS!!! it is ridiculous....

blazing hot AI take that techies are definitely not ready for, caps 

it's hard to imagine that our brains are doing anything much more complex on a basic level though. obviously it's not a 1:1, and the metaphor is NN is like brain, not brain is like NN (the direction there is super important imo), there is something interesting going on.

blazing hot AI take that techies are definitely not ready for, caps 

@thufie but one thing i really really find irksome and dangerous is comparing a system built to industrially put the average prisoner in prison, or put find the average dog, or figure out the average best game of X, etc, calling such narrowly instrumental "minds" intelligent is finally getting to the center of the fact that capital only cares about brains as long as they do work that can make money

kinda dark 

@teh_dude we have finally reduced the worker to the minimum required number of neurons for capital accumulation

kinda dark 

@thufie quite literally lol. "it's too expensive to pay someone to look through security footage, lets replace them with a neural net" "it's too expensive to pay a lawyer, train a net" "it's too expensive to hire an editor, get a neural net"

idk as long as it results in fully automated luxury space communism i'm fine with it, but it would be cool if we expanded the definition of human rather than narrowed it down :/

@thufie

It's even worse than that.

I could remember the details incorrectly, but researchers attempted to train a neural network to identify severe covid from medical imagery.

It instead learned to identify patients who were lying prone, when the images we're taken.
Patients with acute severe covid were just more likely to lie prone...

@void oh yeah this post already assumes they didn't just overfit like 90% of these neural net bandwagon people do and capture the wrong patterns :P

blazing hot AI take that techies are definitely not ready for, caps 

@thufie is this still a hot take after the many publicly documented failures? Real question

blazing hot AI take that techies are definitely not ready for, caps 

@auxeras *apparently* ? yeah i deal with these research people still all the time.... they think it is magic

blazing hot AI take that techies are definitely not ready for, caps 

@thufie really sorry to hear that

Sign in to participate in the conversation
Pixietown

Small server part of the pixie.town infrastructure. Registration is approval-based, and will probably only accept people I know elsewhere or with good motivation.