mediaphage posted:

yeah and then you have thousands of people looking at a grainy computationally processed phone video going YOU IDIOT THATS CLEARLY NONHUMAN

Spergin Morlock posted:

is your friend Linda yaccarino? lol

sharknado slashfic posted:

Dead serious first thing I thought of

sharknado slashfic posted:

Dead serious first thing I thought of

Rudeboy Detective posted:

AI could have given us Iain M. Banks The Culture. Instead we stepped too far ahead of ourselves in an uncontrolled cycle of models eating their own poop and now AI sucks and is bad and very, very lame.

Or something.

Rudeboy Detective posted:

AI could have given us Iain M. Banks The Culture. Instead we stepped too far ahead of ourselves in an uncontrolled cycle of models eating their own poop and now AI sucks and is bad and very, very lame.

Or something.

SniperWoreConverse posted:

but isn't the mp4 off the phone computationally processed by the phone which is a computer i dunno

Rudeboy Detective posted:

AI could have given us Iain M. Banks The Culture. Instead we stepped too far ahead of ourselves in an uncontrolled cycle of models eating their own poop and now AI sucks and is bad and very, very lame.

Or something.

my bony fealty posted:

friends don't let friends go to Subway

euphronius posted:

ah the no true Scotsman defense for ai. a grifter staple defense

euphronius posted:

ok what is AI it is not things everyone says is Ai

Tekne posted:

garbage in garbage out in it's final optimized form

mediaphage posted:

yes that’s my point - that modern lovely smartphone cameras are apply a lot of algorithmic processes so what you capture is not necessarily what’s there

combined with a tiny sensor digitally zooming in on something thousands of feet away and it’s pretty lol.

Squizzle posted:

Hatebag posted:

that's right. god sits in a big magic wheelchair. probably got a toilet in there too

euphronius posted:

ok what is AI it is not things everyone says is Ai

Squizzle posted:

merkaba means chariot. remember when kirk asked why god needed a starship (chariot w a throne) and god was like Because my starship is broken, and kirk fought a rock monster

euphronius posted:

ok what is AI it is not things everyone says is Ai

Hatebag posted:

yep, and then the klingons zapped him with lasers, just like in the bigle

Squizzle posted:

it means love, you baka

my bony fealty posted:

ai is fake, unlike the mummies

Fitzy Fitz posted:

Couldn't you just keep refining the pre-2022 dataset, allowing AI to keep improving but locking it into an artificial timebox? A thousand years from now it could be very advanced but unintelligible as language has shifted, like Latin masses to illiterate medieval audiences.

Azathoth posted:

The problem is that AI of the type we have now doesn't rely on refining existing data but rather taking in massive quantities of data and detecting patterns within that data. Or, put another way, the data is as interpreted as it can be and it needs more data to interpret that existing data further.

The problem is that any data coming in after AI struck big now has content generated from the patterns detected in the training data. The AI being trained on that data will detect the patterns in that data that the other AI put into it and conclude that those patterns represent how humans communicate, but that isn't necessarily true. A small hallucination in the training data can give rise to later, larger hallucinations because the trained AI will replicate the hallucination as valid, which then leads to even larger hallucinations as the AI gets fed increasingly more tainted data.

Also, the problem isn't with AI being able to produce grammatically correct sentences, the existing data allows that just fine. Your solution might be theoretically workable, but it also means that AI is useless for talking about anything that has happened recently, so no writing news stories or discussing current events and be prepared for it to confidently state wrong answers for anything where our understanding has changed. Like, on a factual level it would think that Attenborough's long-beaked echidna is extinct (it was observed last year for the first time since 1961), that Joe Biden is the president, etc.

AI as we have today only gets better than what we have with a constant stream of new data. What we have today isn't particularly useful, and where it does have valuable uses, it doesn't justify the cost. It's why every AI company is burning dump trucks full of VC cash. They need orders of magnitude more data to make their models work but by their very existence they cause less of that to be created. It's possibly an unsolvable problem without a significant breakthrough that changes how the models work.

Azathoth posted:

The problem is that AI of the type we have now doesn't rely on refining existing data but rather taking in massive quantities of data and detecting patterns within that data. Or, put another way, the data is as interpreted as it can be and it needs more data to interpret that existing data further.

The problem is that any data coming in after AI struck big now has content generated from the patterns detected in the training data. The AI being trained on that data will detect the patterns in that data that the other AI put into it and conclude that those patterns represent how humans communicate, but that isn't necessarily true. A small hallucination in the training data can give rise to later, larger hallucinations because the trained AI will replicate the hallucination as valid, which then leads to even larger hallucinations as the AI gets fed increasingly more tainted data.

Also, the problem isn't with AI being able to produce grammatically correct sentences, the existing data allows that just fine. Your solution might be theoretically workable, but it also means that AI is useless for talking about anything that has happened recently, so no writing news stories or discussing current events and be prepared for it to confidently state wrong answers for anything where our understanding has changed. Like, on a factual level it would think that Attenborough's long-beaked echidna is extinct (it was observed last year for the first time since 1961), that Joe Biden is the president, etc.

AI as we have today only gets better than what we have with a constant stream of new data. What we have today isn't particularly useful, and where it does have valuable uses, it doesn't justify the cost. It's why every AI company is burning dump trucks full of VC cash. They need orders of magnitude more data to make their models work but by their very existence they cause less of that to be created. It's possibly an unsolvable problem without a significant breakthrough that changes how the models work.

Rudeboy Detective posted:

What if, just hear me out... one of the mummies was posting in this very thread? Imagine that.

Orbs posted:

Good garbage in (good shitposts that agree with me), no more bad garbage going in (bad shitposts that don't agree with me)