> Tons of content are produced with AI.
> AI gets trained with human + AI mixed information
> More AI content is dumped on the internet
> AI gets trained with more AI produced content....
How does this end?

AI ban

>>16733441 (OP)
AI training off itself isn't actually bad as long as there is a corresponding exponential increase in the human data within the set, so AI is training off itself for now, but once companies run out of access to more organic data they'll be in trouble.

There can be multiple possible solutions. So either we ban it like this guy said:>>16733442 or we make some kind of file property which indicates that the image is ai generated. This would also make it trivial to tell if an image is generated as you could simply look at the file properties. This wouldn't work for language output, which would most likely have to be banned.

>>16733442
if only, more like human ban

>>16733462

>This wouldn't work for language output, which would most likely have to be banned.

Why not? Except for very short texts like "What do you want me to do next?" (which are irrelevant for a training set) you could likely run some math over the text and tell a probability of it being generated, then either exclude above a certain threshold or put lower weight on it. No withstanding that at least" in house" you could likely compare against a log of output texts and simply filter the data (which would not eat up too much resources). Hey, these chatbots got recognizable writing styles, at least for me, so there must be a pattern to be detected here.

>>16733441 (OP)
the fact that using generated training data is known to degrade the quality of a model implicitly distinguishes human-created data from machine-generated data as being superior. i think this should be a bigger talking point in the ethical application of ai.

>>16733441 (OP)
Humans train on human datasets too, it's not like that's a problem.

It ends with extremely low wage human workers creating new content en mass to feed the machines or curating old content. The work will happen behind the scenes, and the finished product will be entirely credited to AI. In the end you will be slaves to the machine you invented to be slaves for you. It will become a tool to further dehumanize people. What did you expect? AI is trained from the works of humanity, so basically it has all of humanities faults, with none of the morality or emotions that correct for those faults.

>>16733441 (OP)
>How does this end?
Model collapse. All dogs are golden retrievers.
https://www.freethink.com/robots-ai/model-collapse-synthetic-data

>>16733441 (OP)
The acceleration of the memetic apocalypse.
https://www.youtube.com/watch?v=zG29C4skboc

>>16733653
This is how democracy ends: with imaginary applause.
https://www.youtube.com/watch?v=pco91kroVgQ

>>16733654
>Memetic engineering isn't real!
>Memetic engineering can't hurt you.

>>16733462
>as long as there is a corresponding exponential increase in the human data
Of great so The System needs YET ANOTHER vector of infinite growth...
>We can sustain this!
>All we need is an infinite amount of work!