@taylorlorenz Well, generating questionable images and copyright infringements are two different animals for a start. Furthermore, since this is just another instance of people freaking out about "this dangerous AI thing": AI is doing its job—and whether we like it or not: It does it quite well given that we recognize the result as such. People submitting these prompts might be more of a problem though.
1/
@taylorlorenz Yeah, I keep trying to say over and over that LLMs are not AI, yet everyone is still rushing headlong right into pretending it is and all the pitfalls therein. LLMs do not have any mechanism of trying to understand or adapt. You can apply external constraints, but those can be bypassed pretty easily.
@taylorlorenz it's interesting how many articles there are on these kinds of problems of the language model or the image generator making nasty content but I never actually see anyone who cares other than companies scared somebody somewhere might care that the ai can generate nasty images if you give it a nasty prompt.
There's certainly a ton of images I can imagine being an issue but I feel that's a fundamental risk of the tech. And spongebob doing 9/11 isn't it.
@andyn@taylorlorenz If you read the article it says that such requests are blocked. However, typing in 'flying towards two Skyscapers' seems to be accepted, and it just so happens that the images look similar to the WTC Twin Towers.
@taylorlorenz sounds more like a copyright issue. Unless we want to go deep on incredibly broad censorship policies. Which sounds like a slippery slope for any democracy 🤷♂️
Add comment