Genuinely, I think the most publicly visible AI advance so far this year is how thoroughly and abruptly the "image generators can't do legible text" problem has beem fixed.
For those who haven't been paying attention, this is a recent ChatGPT output. It can do this sort of dense sensible text in images totally one shot, so don't rely on bad text to identify AI images any more.
Was the text also generated? was it generated before, and given as an input, or was it generated with the image?
The text was its own invention.
Here is the entire interaction that produced this image:
I've seen other people do much more sinister things with this! You can produce 100% realistic fake screenshots of news articles, tweets, that sort of thing, if you can convince the bot it's kosher to do so.
I've seen other people do much more sinister things with this! You can produce 100% realistic fake screenshots of news articles, tweets, that sort of thing, if you can convince the bot it's kosher to do so.
What, like it's hard?
You can manipulate the Document Object Model with Inspect Elements and do this for yourself, you know.












