I'm always shocked by the laziness that goes into made up data. Like here you have multiples of a number. In bio you often see stretched or inverted images that are portrayed as individual replicates, say. You'd think that if you were going to go to the level of just plain cheating that you'd put your back into it. There are statistical tests, e.g., that can tell you with pretty good precision whether numbers are random, say, or whether a large group of numbers is spaced in a way you'd expect a natural set of data to be spaced (e.g., the frequency of small numbers increases with the size of the number). I doubt it would be that hard to fake a set of data if you reverse engineered it to pass the standard battery of smell-test statistics. But you never see that. Or maybe only the dumb ones get caught?
I mean... there's no way to not get caught faking superconductivity. But apparently SpinLaunch and Theranos (and so many others) are allowed to swindle investors for long periods of time. Maybe Dias got inspired. I should also say that obviously a lot of good can come from private industry and profit motives. And occasionally there's public sector flops, like the faster than light neutrinos, for example. Those guys seemed to know they were wrong and just wanted help figuring out why, though. Honestly, I'm having a hard time thinking of why someone would lie if they intend to stay funded by grants.
Yeah, what’s stopping people from using AI generated synthetic data?
Nothing. Can't find it now, but I saw just the other day that someone had got a paper containing the phrase "As a large language model, I can't..." somewhere in the meat of the paper past peer review. Apparently this here comment is the first on hubski to introduce the concept of enshittification. It's not only affecting online or social media platforms, obviously.