As if there weren't already enough ethical concerns with AI art, now this:
AI don't trust techbros
- Thread starter NovaSaber
- Start date
It's a shame, really. AI could improve our lives in amazing ways. Heck, it's already helping improve pathology, but someone who is willing to steal data to improve their AI models would also likely be willing to use said AI to do various other unethical things in the future.
The funny thing is AI is a misnomer, as there's nothing intelligent about it yet, but because it gets called such and no one argues against it, people assume it's like Data in TNG, or will be soon.
An example conversation with the Bing "AI" based on the supposedly amazing ChatGPT(spoilered for length):
Then he finally got it to update it's clock via another question(don't have the image for that one), which led to:
Note: the two question responses in the last image were apparently edited out by a some human moderator between when it was asked and when the poster went back to screenshot.
Also, it's still fairly easy to bypass the filters:
infosec.exchange
Can't embed this one because the gif got turned into a video.
It's all just probability matching of words based on the data it was fed, with no understanding of the actual content. Machine Learning is still the correct term at this point IMO. Art "AI" is the same way, which is why you end up with endless teeth or 6 fingers, because it doesn't understand the concept of what a hand is, just that these pixels are usually nearby when that part of the image is called a hand.
And that totally doesn't even factor in the whole "Garbage in, garbage out" that the filters on ChatGPT try to hide. Or from what I recall, that they had to hire a bunch of cheap "sweatshop" labor to help build the filters it has in the first place.
Also techbros are already trying to screw with the writing industry like they have the freelance art one:
An example conversation with the Bing "AI" based on the supposedly amazing ChatGPT(spoilered for length):
Then he finally got it to update it's clock via another question(don't have the image for that one), which led to:
Note: the two question responses in the last image were apparently edited out by a some human moderator between when it was asked and when the poster went back to screenshot.
Also, it's still fairly easy to bypass the filters:
Marcus Hutchins :verified: (@[email protected])
Attached: 1 image Bing's AI has a list of pre-prompts (essentially rules) that instruct it to avoid discussing certain controversial topics, providing opinions, or saying anything that could be considered hateful. If you are able to bypass the pre-prompts, things gets really, really wild...
Can't embed this one because the gif got turned into a video.
It's all just probability matching of words based on the data it was fed, with no understanding of the actual content. Machine Learning is still the correct term at this point IMO. Art "AI" is the same way, which is why you end up with endless teeth or 6 fingers, because it doesn't understand the concept of what a hand is, just that these pixels are usually nearby when that part of the image is called a hand.
And that totally doesn't even factor in the whole "Garbage in, garbage out" that the filters on ChatGPT try to hide. Or from what I recall, that they had to hire a bunch of cheap "sweatshop" labor to help build the filters it has in the first place.
Also techbros are already trying to screw with the writing industry like they have the freelance art one:
A Revolutionary Tool Gives Artists a New Weapon in the Fight Against AI Art Theft | Artnet News
A new software developed by computer scientists aims to protect artists whose work is at risk of being used to train AI to copy their style.
news.artnet.com
This looks promising. It's a filter that subtly alters specific details in an image just enough in just the right ways to confuse generative AI models, while remaining imperceptible to a human viewer. The best part though is that if these filtered images are mixed in with unfiltered ones, it's almost as effective.
Last edited:
Ford files patent to allow self-driving cars to drive away from owners who don't keep up with payments
Repossessions are about to get a whole lot more impersonal.
People's futures are being put in the hands of machines that translate without comprehension.
So now Google's making the first serious push for the audio equivalent of all those AI image generators:
blog.google
I could see record companies fighting this hard, on the basis of what's in the training set, or I could see them jumping at it to basically auto-generate boy bands and pop idol vocalroids effectively - like Hatsune Miku or Sharon Apple, but without having to pay anyone or actually do the writing beyond whatever the licensing cost is. Or even split on it, with some going one way and some going the other.
Turn ideas into music with MusicLM
Starting today, you can sign up to try MusicLM, a new experimental AI tool that can turn your text descriptions into music.
I could see record companies fighting this hard, on the basis of what's in the training set, or I could see them jumping at it to basically auto-generate boy bands and pop idol vocalroids effectively - like Hatsune Miku or Sharon Apple, but without having to pay anyone or actually do the writing beyond whatever the licensing cost is. Or even split on it, with some going one way and some going the other.
Then they'll just throw money at Congress until the law's changed.
Hahahahahahahaha
Things like this are why I'm not even slightly concerned about any of these machine learning apps.
I was honestly half expecting it to be one of Trump's lawyers.
You can't synthesize stupidity.