Predictions about the impact of AI-generated imagery

One of the people running chatgpt says it actually costs multiple cents to run a query but they're trying to get it down

This is where the rubber meets the road, for starters: https://twitter.com/tsrandall/status...974341632?s=20

This guy used chatgpt to write a sorting script for an excel spreadsheet. He's way out of his element but was able to consult it in plain language like a tutor. He didn't feed it a prompt, he had a conversation with it. And when the code ran into errors, he fed those back into the program for troubleshooting advice

I've tried using dalle2 to outpaint some archival photos that were portrait orientation and required tighter than ideal cropping to fill a 16x9 frame. The results were unusable, but it's possible that I'm using the wrong AI. I may experiment with stable diffusion. This could be very useful for me very soon
 
A friend of a (new) friend asked ChatGPT to write an article on central planning. And so it did. And it came out pretty good'ly.
 
So this ChatGPT is acting more like a personal assistant at this point. I could see this being for hire as the tech improves just like you would hire a software programmer. If it is trainable line an employee it brings value.
 
So this ChatGPT is acting more like a personal assistant at this point. I could see this being for hire as the tech improves just like you would hire a software programmer. If it is trainable line an employee it brings value.

I assume that there are means of automating it to do larger, more complex tasks. Which is still like a personal assistant perhaps, except they might be able to take some of the tasks you're currently handling

The risk is that the program is sometimes wrong, so you need to be careful to doublecheck its work and not overly rely on it

I used it last night to help me brainstorm ideas for archival b-roll for an edit, things like 20 iconic images that represent Americana, the American west, the American Dream. I would have done the same thing using Google and probably arrived at the same answers, although this was faster.

I asked it for ideas for elegant transitions between one image and another (looking for visual parallels, etc, places I could draw visual analogies) and it just told me I should use fades because they're elegant. No help there
 
Check out 'mTransition Ink 2' or 'mTransition Shade' or some of the other ones on this page...my go-tos:

https://www.motionvfx.com/store/fina...sitions&page=1 (this company makes tons of wedding stuff, especially templates)

Thanks, that could be useful, but I was thinking more things like a horse becomes a train becomes a car which is kind of what I did. Or the blue of river becomes the blue of a sky, the Statue of Liberty becomes a person in some setting that sort of thing
 
It's just cuts and fades, not actual transformations. But actual transformations would be cool. I'm working with archival footage and just focusing on the sequencing
 
Worth recalling that self-driving technology at one point was making massive and shocking leaps that defied prior expectations and lead many (including me) to expect commercialization much sooner than it actually happened.
Another example is robotics with Boston Dynamics shocking videos

https:/twitter.com/ModeledBehavior/status/1600886657192165376?s=20&t=Y_vlb18GCjwJpknDt_h8JA
 
This is crazy. Here's a laundry list of complaints about the output from the wildly popular avatar generation AI from LensAI.

Asian users say that the output doesn't look like them but rather a variety of random Asian people. Women say that their avatars are hypersexualized and often show their naked breasts (often enlarged) even if they only upload headshots. And apparently the mangled remains of real artists' signatures sometimes get copied in as well, remnants of the original works that are being plagiarized.

basically, the biases of the source material are reflected in the AI"s output. And since it was trained on unfiltered images from the internet (probably primarily American and/or European internet by the sounds of it), it ends up turning women's portraits into porn and being less able to distinguish between Asian people

https://petapixel.com/2022/12/08/us...selfie-generator-is-sexualizing-their-photos/
 
This is odd. Apparently these chatbots just make **** up sometimes.

“What it gives you is kind of like an Aaron Sorkin movie,” he said. Mr. Sorkin wrote “The Social Network,” a movie often criticized for stretching the truth about the origin of Facebook. “Parts of it will be true, and parts will not be true.”
He recently asked both LaMDA and ChatGPT to chat with him as if it were Mark Twain. When he asked LaMDA, it soon described a meeting between Twain and Levi Strauss, and said the writer had worked for the bluejeans mogul while living in San Francisco in the mid-1800s. It seemed true. But it was not. Twain and Strauss lived in San Francisco at the same time, but they never worked together.

Scientists call that problem “hallucination.” Much like a good storyteller, chat bots have a way of taking what they have learned and reshaping it into something new — with no regard for whether it is true.

https://www.nytimes.com/2022/12/10/technology/ai-chat-bot-chatgpt.html
 
This is odd. Apparently these chatbots just make **** up sometimes.



https://www.nytimes.com/2022/12/10/technology/ai-chat-bot-chatgpt.html

Factions of scum***** the world over funding propaganda & misinformation campaigns must be particularly gleeful at the prospect of having dirt-cheap robots invent ever-more horrible distortions than their most cynical human actors. These systems have a way of leveraging the work of so many people and funneling it to serve existing power structures so well -- weird fluke of the technology. Total coincidence that it's all unraveled out to the public through unregulated 'non-profit' entities (in the west at least).
 
Back
Top