Soon people discovered that Meta’s ghoulish posters had been among us for months, even years. There’s Liv, a “Proud black queer momma of 2 & truth-teller,” according to its Instagram profile. Add to that Brian, “everybody’s grandpa;” Jade, “your girl for all things hip-hop;” and Carter, a “relationship coach.” I’m sure there are more yet to be discovered.
All four of these posters have pages on both Facebook and Instagram with mirrored content and all four have post histories that go back to September 26, 2023. The accounts have the blue verified check marks and a label indicating that they’re an AI “managed by Meta.” Users can block them on Facebook, but not on Instagram. Users can also message them across all of Meta’s platforms, including WhatsApp.
Hey Liv! Sup! Jow menee ees and uus in jujubes?
After it replied with a non answer, here we go:
As you can see, this is still a great way to out an AI.
Great.
Na weeg unna haf tamayka brannu dielek jussa fak widda sinthoz
Now we’re gonna have to make a brand new dialect just to fuck with the synths
God damn that is fucking hard to read.
Someone’s never taken a linguistics 101 course 😜
I have not. But english is also not my native language so my brain reads that gibberish in a different way.
w3 c@n @1w@y5 60 b@ck 2 1337
I can instantly read that, and I’m proud to have been alive during 1337 speak. Thank you, friend, for the awesome memories
1337 43V@
Bots probably already know that though because of what they’ve been trained on. Actually … Let’s see Gemini do it for us.
Not too bad.
The inverse is more impressive to me.
But it can be defeated by trying slightly harder. Multiple characters, inconsistencies, adding slang and loanwords, the same tricks we’ve always used to get around censors and oversight.
Of course we’d have to keep it out of their training data and constantly change it… cyberpunk future is exhausting already.