Doctor Mike

And that's how you can tell that certain clusters of accounts are not human operated in that moment.

Sometimes accounts are operated at times by a human and at times by a machine.

But when you look at some of these slips and then you look at what the account posts about when it is not slipping.

So a slip might be as a large language model, I cannot.

That was a common phrase that ChatGPT or OpenAI's API would return.

That actually became kind of a meme, so that one became useless because other people started saying it.

As a large language model, I cannot became something that... Yeah, which makes it harder to study.

But this is what I mean, right?

When you're studying, like, this is why you can't just rely on the content because then you'd be like, oh, well, anybody who says as a large language model, I cannot is a bot.

And then you're going to have a ton of false positives because anybody who uses it as a meme jokingly, you're going to misidentify as a bot.

Yeah, no, I was in a pro-vaccine Facebook group and it was demoting and penalizing accounts that were sharing anti-vaccine links.

But sometimes it was doctors in the group who were throwing in the links saying like, oh, somebody should write a response to this.

And so you had to come up with new and novel ways to get around the auto mods to make sure your group didn't get penalized like...

Yeah, I mean, these are the, again, this is the sort of like the cat and mouse game of, you know, you would see, I remember during COVID, people would write the word vaccine using like two apples for the C's, like emoji apples or like two skulls, whatever.

You know, the kind of futility of automods and vaccines.

I mean, this is where, you know, you saw Zuckerberg say this and I think it was January where he says there's too many false positives.

There's, you know, it's like precision versus recall is how it's described in machine learning when you're training a system.

Do you.

what is the threshold that you want to set your system to?

Do you care more about false positives or false negatives?