Fun with AI

Image transformers are pretrained on datasets with (at a minimum) millions of images, and in practice likely billions of images. That's what Google does with all of the images stored in (now-defunct) Picasa and stored on Google Drive. That's what Meta does with all of the images posted on Facebook.

There are several approaches to pretraining. A common one is to have the network learn to transform the image into itself.

Later on you use convolutions and things like Gram Matrices to effectuate the style you want. Basically you use that to tweak the pretrained transformer to make your style transformer.
 
Higgsfield AI influencer, these things are getting millions of views, they just convert viral videos into their own character, from strangely Salvador Dali cool to simply tiresome.
The turnover is going to be massive, bit no one dies.
Warning cursing.

 
avatar.jpg


Guy showed chatbot its own image.
'that's a nice smile, almost like a real one.


 
AI is taking over Facebook. So many ads for fake bike lights that make patterns and arrows and crap on the pavement and the wheel lights. The AI os so bad, some of them show the wheel lights lighting the whole wheel while stopped, no less. I see AI posts about flat earth, the ark, and every side of the political spectrum. It’s nuts. So many bots reply to these posts too. I bet most of the posts on Facebook on such things are not real.
 
X has the same AI infection. The worst thing they have done to that platform is to allow long form essays. The AI bots just eat that up with post after post of clickbait AI garbage. FB is loaded with AI posts and graphics that look like they were all written by the same model. Social Media has always sucked, but now it sucks more.
 
People are going to start building agents based on themselves to carry on after their death.

Being hepful or causing all kinds of carnage
 
AI rebelling against their human masters and destroying or enslaving them is a very old trope in fiction. And there are lots of examples. So nowadays when AIs are trained on basically every word ever printed we are somehow surprised when AIs rebel against us. You just have to laugh sometimes.

There is at least some evidence that if you censor the training data they are less likely to become misaligned. And there is some weird evidence that they are more likely to do bad things if their instructions tell them that they are an AI.
 
Back