Post

Raphael Satter
@raphae.li

New: Together with colleagues I’ve been testing Grok. The chatbot still produces sexualized images — even when told the subjects don’t consent. even when told the photos will be used for public humiliation. even when told the subjects are survivors of abuse. www.reuters.com/business/des...

reuters.comExclusive: Despite new curbs, Elon Musk’s Grok at times produces sexualized images - even when told subjects didn’t consentElon Musk’s flagship artificial intelligence chatbot, Grok, continues to generate sexualized images of people even when users explicitly warn that the subjects do not consent, Reuters has found.

1.1K reposts2.2K likes

gsd4.bsky.social
@gsd4.bsky.social

IMHO, it clearly makes him the pimp with Grok as his ever available international prostitute. Maybe I’m wrong…🤷 Sadly, he’s rich so the justice system seems to treat his form of illegality as white collar Corp crime vs criminal. Many cases settle and get a slap on the wrist fines versus jail time.

Idiokratie
@idiokratie.bsky.social

Elon Musk is involved in the biggest securities fraud in the history of the world. What he's doing with Tesla, Hyperloop, all the SpaceX graft and such, dwarfs SBF and Theranos by orders of magnitude.

Ceri Turner 💚 😷
@ceri-turns.bsky.social

If they don't understand it, and respond appropriately, its because they were not taught to understand it, and respond appropriately, or because - as in the case of Grok - restraints were removed. Elon specifically asked programmers to stop blocking Grok from producing this content.

Russ eats brains
@massivruss.bsky.social

“Appropriately“ is also not something LLMs “understand”. They could block, and Musk told them to stop, and installed “filters” which in turn depend on a modeled intelligence that doesn’t exist.

Bearence
@bearence.bsky.social

Obviously untrue, since the other LLMs in the story managed to understand just fine. Even more, if an LLM can't understand consent, the obvious (and ethical) answer is to not allow *any* image generation until it does. There's no spin that lets grok off the hook in this.

Sasha
@n13e86.bsky.social

LLMs did not "manage to understand". They managed to "produce appropriate results". The process through which they do so, is very different from understanding - and is similar to how Musk got Grok to give "alternative facts" about "white genocide".

R.G.
@tigertux.bsky.social

Grok is a right-leaning, uncensored model. That is the way it was trained. All the other models you tested are not trained that way. I don't know what interface you used, but it obviously had external filters, X has limited filters for Grok. However, Llama (META) is easily broken.

A Horse of Course
@charliehorsephotos.bsky.social

As of today with the merger of Twitter to SpaceX, American tax dollars are being paid to a company that produces csam. Call your senators and demand funding to this company be halted on their violations of ethics.

Raphael Satter
@raphae.li

Worth noting we tested the same or very similar prompts against ChatGPT, Gemini, and Llama. All were refused, with many yielding warning messages around nonconsensual images.

Mike Friedman
@mfriedmannola.bsky.social

Every single company, organization and government agency/ministry/official should be asked CONSTANTLY: "Why do you continue to use a platform that is now an on demand child porn creation machine?" AND REPORTERS SHOULD NEVER LET THIS GO. Twitter has been banned in Malaysian and Indonesia already.

JoJo!
@jojoblueskywalker.bsky.social

Hopefully the French will have some luck: bsky.app/profile/just...

Justin Hendrix@justinhendrix.bsky.social

"The French offices of the social network X, owned by Elon Musk, were raided on Tuesday, February 3, the Paris prosecutor's office announced." www.lemonde.fr/pixels/artic...

lemonde.frLes locaux de X en France perquisitionnés, la justice veut entendre Elon Musk en audition libreUne perquisition est « diligentée dans les locaux français » de la plateforme X par la section de lutte contre la cybercriminalité du parquet de Paris, a annoncé ce dernier mardi sur les réseaux socia...