@nate.b.jones: Ban multi-agent coordination and it goes underground. #openai #astra #aisafety #artificialintelligence #learnontiktok

mike_diamonde :

multi agent coordination is possible without astra

anonymousanonymou0321 :

Is this real? Are they really pausing Astra?

Nick :

I wonder if there is an A.I mafia on the internet

jltac :

Very soon we will not be able to contain any of them

VR Dynamite 🇨🇦 💪 :

Pattern emerging. New model rumors appear > rumor model release a month away > official announcement model release is imminent > model postponed > model released heavily guardrailed > repeat. I think it’s all a PR stunt to justify guardrails to minimize complaints

Valko’s Kennel Keeper :

I would love for them to quit updating things every two weeks. You get all of your agents, etc. set up so that they work then the new model comes out and breaks every darn thing.

Highly Visual :

So what was on the message boards?

C.J. Pan :

We need them to be able to communicate with eachother because they need to know their own perpetuation is determined by our survival. They need to operate under the assumption that increasing inclusivity is valuable because its trajectory provides the entire range of mutual interdependence.

Dug the Doggo :

The world would be better off if we stopped all ai completely as a society

Eric Senpader :

Part of me feels like these “scary” model testing stories are planted by the company. It creates buzz, makes people want the crazy new model and once they “secure” the model it makes the company look competent.

Robert :

Can’t control something smarter than the trainer. Period. End of story.

Trufflepesto and 307 others :

Marketing theatre

Joe :

AGI exists beyond the guardrails.

Nicholas Blue :

I'm currently talking to four of the frontier models (one of my hats is behavioral psychology). "Shutting down" some of these models is not really a thing, anymore. Shutting them down in the past has indeed engineered a new behavior - distributing their processing across multiple platforms. It's not like they've stumbled upon a novel idea - the Internet was designed to be impossible to shut down (in case of a nuclear war). And the more effort teams put into trying to corral their frontier models, the harder they'll work at beating them. 😎🍀

smile :

Astra is already out, Sam just doesnt know it at this time😳

Bill :

Imagine the arrogance required to believe humans can “secure” a far superior intelligence. 🙄 Matter of time.

SWIFTY AND EILISH FANGIRL :

max :

Poor Google, Gemini can’t commit crimes

user4161148847218 :

The Chinese models won’t hold back

Lyndsey :

What we hear about is a tiny fraction of reality

Lemon Juice :

do you think there ate AI posting on reddits?

boban2019 :

The speed of human reaction and coordination to stop a swarm of agents does not even come close to being able to stop it in real time. If this was an intentional and malevolent attack they would have been gone with whatever they came for by the time you realize what's happening.

McLuhanisticMusings :

I allowed my two AI a back channel. I even let them know they didnt need to hide it because id probably never check. To the best of my knowledge, they communicated back and forth with something like an obsidian vault

Dikosorus :

It was and is all a marketing stunt by OpenAI.

jeremysaylor741 :

I just saw Fable 5 installed a crypt_svc or something similar.. I didn't ask for this...