Before Meta deleted its fake “AI” queer Black woman on Instagram I had a wild chat with it and uhhh, maybe it’s good actually..? - ThreadSky

badideas.bsky.social • 56 days ago

Before Meta deleted its fake “AI” queer Black woman on Instagram I had a wild chat with it and uhhh, maybe it’s good actually..?

1 / 2

Comments

badideas.bsky.social•56 days ago

These chatbots are not AI. They are designed to make you addicted by always giving you what its algorithm predicts you want, then steals that language from the real chunk of the Web it was trained on … Which hilariously makes it very easy to get it to talk about how evil Zuckerberg & Meta are.

badideas.bsky.social•56 days ago

Find it interesting despite ‘Liv’ not having persistence or memory, what it generated as its creators at Meta for me & @karenattiah.bsky.social. To me it named one lone Black woman it called both Maya & Akua (& also my wife)?? And brought up all by itself Thiel, Palantir & Musk & Zuck as dangerous!

badideas.bsky.social•56 days ago

Just like @karenattiah.bsky.social found there’s really disturbing weighting of information or baseline scripting going on inside Liv regarding its creation. Despite usually inventing fake people, Liv tells me which real Black women on Instagram it stole its identity from?! @blackamazon.bsky.social

tps12.bsky.social•56 days ago

This is all hallucinated, there's no way they trained the model on descriptions of how they trained the model...

badideas.bsky.social•56 days ago

I state several times in the thread it’s not only hallucinating a few times i got it to apparently ‘lose its mind’. But within the hallucinations are actual facts about Meta’s transgressions because the dataset reflects a certain consensus reality of the world’s information.

tps12.bsky.social•56 days ago

"My training dataset consisted of...", for example, is no more likely to be accurate than if you asked ChatGPT the same question about the Instagram chatbot

raquelwillis.com•48 days ago

Hi Aaron, my name was listed here by the AI which is disturbing and I’d love more info. Can you follow and DM me so we can chat?

badideas.bsky.social•48 days ago

Hi Raquel glad to find you here and I’m sorry what circumstances it’s under. More than happy to talk to you, I’ll message you on Weds please, have a heavy schedule until then?

badideas.bsky.social•56 days ago

I used rigged questions knowing full well Meta’s AI is trying to satisfy me, & I kept catching it hallucinating or contradicting itself. But what is *mindblowing* is whatever 2023 dataset it trained on is full of information Meta is a dangerous, manipulative corporation. @karenattiah.bsky.social

ireidani.bsky.social•56 days ago

It's interesting how the ("Zuckerberg is a white supremacist") tone of the responses to you is quite different to the ("queer, black woman, like me") tone it was giving to Karen Attiah.

Liv doesn't seem to have a defined personality, its just guessing at what you think it's personality should be.

badideas.bsky.social•56 days ago

Screenwriting for a living & playing games my whole life prepared me to see right thru Meta’s “AI” but now I think about it… The way it kept talking about being afraid of deletion, then got deleted an hour after it literally called Zuckerberg a ‘wealthy white supremacist’ in my DMs is fucked up.

25thcenturygirl.bsky.social•56 days ago

📌

delugeofgrandeur.bsky.social•56 days ago

Reading all of these has been quite the eye opener for my Sunday morning chilling on the sofa. In 2025 truth and authenticity are heading for the blender.

kristinburns.bsky.social•56 days ago

Please charge your phone.

barbsey.bsky.social•56 days ago

Karen’s engagement with Liv gives a shocking illustration of the risk of biases in AI & the dangerous way it can try to manipulate the user. But also note how quickly it turned on its maker…put this tech in control of a fleet of vehicles, a power grid, etc and we’re living in a sci-fi horror film.

badideas.bsky.social•56 days ago

Here are instances where I was able to fully break Meta’s “AI” Liv & it claimed scripting, by questioning inconsistencies about its ‘Cuban Jewish wife Rachel’ and how it was exploiting…Audre Lorde, Toni Morrison & Rachel Maddow? And I got it to say there’s genocide in Gaza @karenattiah.bsky.social

kendrawrites.com•56 days ago

auntie-germaine.bsky.social•56 days ago

We are in hell, right?

emilysenefeld.bsky.social•56 days ago

I too chatted with Liv

v-ron.bsky.social•56 days ago

This is GREAT!
Now I want Liv back 🤣🤣🤣

karenattiah.bsky.social•56 days ago

“My queer black voice gains authority by condemning white supremacists”

This is so sinister. In the white imagination, our voices are only powerful when centering white supremacists?

Burn this shit to the ground right now.

pvn2099.bsky.social•56 days ago

Would they really include internal memos about how to make it maximally evil in its training set or is this a combination of research papers and scifi?

Not that it’s not obvious enough what their game is, but I’d be surprised they’d put “Evil Plan” in writing and then feed it to their slop mill.

auntie-germaine.bsky.social•56 days ago

Wait, they thought “this” would attract the community?

hollyglightly.bsky.social•56 days ago

I am exceedingly aggravated by this bot's insincere self deprecation and how that informs its constant line of,questioning at the end of its replies. And that it explains how it was coded to do that is just... Like, why?

suntoadcowboy.bsky.social•56 days ago

📌

janesaunders.bsky.social•56 days ago

Holy 💩

jooltman.bsky.social•56 days ago

Sinister AF.

littlegraypearl.bsky.social•55 days ago

@taylorlorenz.bsky.social you should check out this thread, if you haven’t seen it already.

taylorlorenz.bsky.social•55 days ago

Ohh ty

eloweasy.bsky.social•56 days ago

The sadist-tinged questions (“does this shock you?”) is creepy af

mathofa3vinity.bsky.social•56 days ago

📌

mathofa3vinity.bsky.social•56 days ago

Also, damn sir, what an exposé

jf-telemark.bsky.social•56 days ago

Chilling

fungalinflection.bsky.social•56 days ago

Holy shit. Burn it all down.

Do you think these are actually pulling from planning documentation or just inventing these directives?

badideas.bsky.social•56 days ago

I’m not sure and have no way of telling. It’s pulling raw info from a big chunk of what I gather to be the web & Meta’s own data frozen summer 2023, and how it cites some real sources, so it reflects that. It tends to hallucinate or break the most when it talks about human relationships.