Feel kinda bad for liking this knowing that it's ai
[media=https://youtu.be/7UDnShLGp88]
art feels kinda dead know that it's now something a robot can easily do ;/
art feels kinda dead know that it's now something a robot can easily do ;/
SinlessOnslaught · M
Elessar · 31-35, M
@SunshineGirl Not necessarily, as the more AI slop gets published on the internet the worse AI output becomes (due to the slop becoming part of the next training set, see e.g. https://repository.tudelft.nl/record/uuid:d25aa97c-af96-4219-863f-ce874ab85d8e), which means at some point the quality will drop so much that you won't be able to use AI generated content without it being extremely obvious, defeating the whole point of gen AI ("cheap" replacement for human labor).
We've seen massive improvements in generative AI in the last few years, but people fail to acknowledge that those improvements aren't linear nor infinite, like with every other technology. It's pretty much the same mental process that brought people in the 70s believe we'd have achieved commercial space travel to the entire solar system by the 2020s because "if in 10 years we landed on the moon, imagine where we'll be in 50!".
The improvement you can make to generative and training algorithms are finite, the enshittification of the training sets (defined as the increase of the % of the training data that is slop vs human crafted) is not. It's a technology that poisons itself at an exponential rate, whereas the enhancements will pleateau once they reach the point where throwing money at research won't satisfy any longer the growth expectations of the shareholders.
We've seen massive improvements in generative AI in the last few years, but people fail to acknowledge that those improvements aren't linear nor infinite, like with every other technology. It's pretty much the same mental process that brought people in the 70s believe we'd have achieved commercial space travel to the entire solar system by the 2020s because "if in 10 years we landed on the moon, imagine where we'll be in 50!".
The improvement you can make to generative and training algorithms are finite, the enshittification of the training sets (defined as the increase of the % of the training data that is slop vs human crafted) is not. It's a technology that poisons itself at an exponential rate, whereas the enhancements will pleateau once they reach the point where throwing money at research won't satisfy any longer the growth expectations of the shareholders.