Hacker News Reader

Hacker News

Comment by simoneau | Hacker News Reader

Full thread

simoneau

Me: Facebook AI, please post an entry about my vacation on Cape Cod and create a bunch of photos to go with it.

Facebook: Great. I'd be happy to. Any more detail you'd like to add?

Me: Make us look attractive. Show that we're a having a great time. Also, we went to see the Chatham Lighthouse.

Facebook: OK, done!

...

Facebook: You've received 48 likes. Your mother would like to know if you had any salt water taffy.

Me: Yes, and please create a picture of my oldest daughter having trouble chewing it.

Facebook: Done.

View on HN

When you think about it, the only thing that's weird about this hypothetical conversation is the context of it being about (purported) photographs.

We expect images that look like photographs — at least when taken by amateurs — to be the result of a documentary process, rather than an artistic one. They might be slightly filtered or airbrushed, but they won't be put together from whole cloth.

But amateur photography is actually the outlier, in the history of "capturing memories"!

If you imagine yourself before the invention of photography, describing your vacation to an illustrator you're commissioning to create a some woodblock-print artwork for a set of christmas cards you're having made up, the conversation you've laid out here is exactly how things would go. They'd ask you to recount what you saw, do a sketch, and then you'd give feedback and iterate together with them, to get a final visual down that reflects things the way you remember them, rather than the way they were, per se.

This is an interesting point. Usually people claim technology goes inexorably forward, yet here we are, merrily destroying trust in the most objective method we have to record the past!

Photographs haven't been able to be trusted since almost the beginning. Trusted as an image of a real scene that is.

Indeed, people viewing photographs have always been able to be manipulated by presentation as fact something that is not true -- you dress up smart, in borrowed clothes, when you're really poor; you stand with a person you don't know to indicate association; you get photographed with a dead person as if they're alive; you use a back drop or set; et cetera.

These aren’t even remotely comparable to AI photo manipulation.

Agreed. My point was that trusting images ('seeing is believing') has always been at issue whilst we might imagine it is a new thing, the scale of the issue is different -- phenomenally so -- but it's not a category difference. Many people were convinced by the fairy hoaxes based on image manipulation in the early 20th Century (~1917). They fell for it hook-line-and-sinker, images made with ML weren't needed.

This is a bit of a stretch, but the end results from either manipulation technique would be comparable if they were meant to skew the truth the same way. However, that sounds stupid as shit when I read it back, but I'm not entirely sure why.

I think a use case for AI image manipulation could be more like if I need a picture where I'm poor but wearing smart borrowed clothes, standing with an unassociated associate and a dead alive, with a backdrop, with the only source image beimg selfie of someone else that incidentally caught half of me way in the background

The intent or use cases for these two (lacking a better term) manipulators aren't orthogonal here. The purpose of AI image generation is, well, images generated by AI. It could technically generate images that misrepresent info, but that's more of a side effect reached in a totally different way than staging a scene in an actual photo. It seems like using manipulation to stage misleading photos would be used primarily for the purpose of deceptive activities or subversive fuckery.

https://web.archive.org/web/20140222103103/http://subterrane...

Sounds like the plot line to an episode of Black Mirror, but also something that is far too likely to happen.

me: Facebook AI, please post a tender moment between me and my father when I was a boy. Include some photos.

Facebook: I'd be happy to. Are there any more details you'd like to include?

me: Please show how he didn't understand me at first, but then he looks at me and starts crying with love and regret.

Facebook: Done. Your relationship with your father must have been deeply fulfilling.

https://petapixel.com/2022/12/14/man-fakes-an-entire-month-o...

This is the weirdest video I ever watched. It's like Black Mirror ... but in real life ... and a somewhat happy ending.

Incredible. Man, am I going to be telling my grandkids about a time when you could believe your eyes and ears on the internet.

What if we're already living in the future, and everything we're experiencing right-now is being AI generated?

...that, and other thoughts I have while baked.

You guys are very unambitious.

FB AI, make a series of posts about me climbing mount everest, meeting dalai lama, curing cancer, bringing peace to ukraine, changing my name to Melon Tusk, announcing running for president and adopting a dog named Molly

But see, that's the sort of thing that would give it away.

You got to shoot for something just attainable enough to sound credible, while still being at the "enviable" end of the spectrum.

"FB AI, make a series of pictures of my first 3 months at Goldman Sachs in 2021. Include me shaking hands with the VP of software as I receive a productivity award for making them $1m in a week. Include a group photo of me and 12 other people (all C execs and my VP must be there). Crosspost all to LinkedIn, with notifications muted."

"Ok done"

"ChatGPT, take my existing CV and replace entries from 2021 onwards with a job as Head of Performance Monitoring at Goldman Sachs, reporting to VP of software. Include several projects with direct CEO and CFO involvement. Crosspost changes to LinkedIn."

"Ok done"

... and now I can go job-hunting.

I can see AI Consulting to be the next incarnation of social media expert

Reply on news.ycombinator.com