One of the early popular examples was asking ChatGPT what the downsides are of generics in TypeScript, and it cited that they're not compatible with some browsers/runtimes. It's entirely believable, but entirely wrong, and actually quite hard to confirm either way with internet searches. It's the sort of thing people would readily believe and pass around, though.
Similarly, when I was trying to concatenate video in ffmpeg, it produced convincing-seeming command lines, almost none of which worked properly, but almost all of which used actual ffmpeg command line parameters with appropriate-seeming formats and names. I wasted significant time on debugging those before realising quite how wrong ChatGPT was.
That's an interesting take, and one I haven't heard before. My main squeamishness with bad journalism, and even worse, with chat gpt, is not truth-content, but rather sourcing.
If you know the source, you can account for its biases, or, your reader can do so. As such, well-sourced quotes of complete nonsense tend towards adding to the sum total of truth in the world, because it was true that somebody said it, and their perspective is part of what's interesting in whatever event you're discussing.
If you scrub the data of its sources (like journalists do routinely, and chatgpt does by necessity) important information is lost.
Any bias is everyone’s bias. Not saying it doesn’t happy at all today . But wow, scary
We will look at these things as devine correctness, and eventually forget their human construction entirely.
Remember how people use to criticize the newspapers? They'd say they read an article on a subject they new something about and were shocked at how bad it was. Then they'd conclude that the rest of the news was just as bad. What wasn't acknowledged was the utility of having imperfect information.