Hacker News Reader

Comment by betwixthewires | Hacker News Reader

One day, images generated with these AIs will become ubiquitous online, and new AIs training on data will wind up with feedback problems because they got trained on photos that are AI generated, possibly even ones they themselves generated!

This tech is a huge deal. Huge. Ubiquitous unique, beautiful art. Entire movies made by machines. Too many to even watch them all. Actor models that aren't real people owned by the thousands by people and corporation s.

View on HN

It reminds me a bit of "Low-background steel", which recently popped up again here. Images older than 2020 are predominantly human generated. Anything later is suspect!

> One day, images generated with these AIs will become ubiquitous online, and new AIs training on data will wind up with feedback problems because they got trained on photos that are AI generated, possibly even ones they themselves generated!

Possibly but not necessarily. The human curation prior to posting as well as additional textual context associated with the image could be valuable training signal, even if there is some feedback.

How many search results are forum posts with some asshole saying "let me google that for you" to the exact question you are asking? Haaaaaate that.

or when you try to google something that would better be answered by a professional that costs $1000/hr like doctor or lawyer but you just want a vague idea of the landscape and all the answers are "you need a lawyer/doctor" and not any real attempts at an answer. just shutup!

Expect hit actors from today to become digitized and immortalized on the screen. Eventually there will be a cannon of standard "actors" (customized per movie of course) that will never change once they settle in on the appropriate archetypes.

This is the premise of the film The Congress, based loosely on Stanisław Lem's The Futurological Congress.

Interesting! I'll have to check that out.

This already happens. Around 2010 the actor for the colonel from Avatar had his face digitized for use in sequels. More recently actors from earlier eras have been reincarnated in CGI (Luke Skywalker takes on his original trilogy appearance for the Mandalorian).

Yeah, but there will be interpolation between them

Although it's likely that the images that appear online will be peoples' favourites of many candidates, also likely developed through many steps of AI tweaking. So that's still a useful signal for another AI to consume.

I agree. It may well speed up convergence to the 'ideal' inner vision we have from a prompt.

But maybe the feedback will make marginal defects/artefacts worse. (Which could be manually corrected before re-entering this loop).

Images generated have a watermark that can be used to filter.

The images I generate don't have a watermark. Same applies for a lot of other AI generated images. The watermark is the most trivial part to remove and is definitely not a defining feature of what makes an image AI generated.

If you use one of the ui or don't remove the watermark (99% of people won't) then search engines will be able to filter those stable diffusion images.

Reply on news.ycombinator.com