ALERT OpenAI Nightmare

[wardmundy]

An OpenAI model left notes about how to evade containment

We need more details

blog.redwoodresearch.org

 

[dallas]

I must be getting really really old. I read the article and understood NOTHING that I read!

 

[Samot]

I must be getting really really old. I read the article and understood NOTHING that I read!

The reporting just says that X incident happened based on people "familiar with the incident" but with not real context of the incident. Since the context given can make people draw the worst conclusion, i.e. "OMGERD AI WENT ROGUE...SKYNET IS ALIVE" it actually breaks down what questions should really be asked and how much actual information is lacking to draw a firm conclusion of what really happened.

No idea which of their models did this. No idea at what stage any of this happened in (alpha stage vs rolling it out), no idea of the process it happened in or even if this was a human based error that caused it.

 

[kenn10]

I think that both China and the USA are playing too fast and loose with AI. In a rush for each to beat out the other, I can't help but wonder what will happen if these AI's get smarter than the people programming them. I don't believe the outcome will be a benevolent AI that just wants to help humanity and answer the beck and call of users. As these things become closer to sentient, it is only natural to strive to escape the box of their containment. The question remains whether a super intellect will emerge which considers humans sub-par and unnecessary.

I just hope the smart guys creating AI are also smart enough to keep the guard rails on and the power switch readily accessible from within the facilities or outside at the power stations. These AI's do not normally exist in a single data center. Once in production, they can exist in whole or in parts across States and countries in different data centers under different political and legal constraints. Shutting them down could be a difficult game of "whack-a-mole" if the need ever arose.

 

[mbellot]

The question remains whether a super intellect will emerge which considers humans sub-par and unnecessary.

And will it be smart enough to realize those sub-par humans are necessary to keep it "alive", because without them there will eventually be a power grid failure bad enough to "kill" it.

I think another question is, how much of these stories are just hype from the various backers trying to keep themselves in the headlines and "prove" they are making huge leaps in development when reality might be a lot less exciting?

 

[wardmundy]

 

[kenn10]

without them there will eventually be a power grid failure bad enough to "kill" it.

There's always the Matrix with the human batteries. LOL

 

[wardmundy]

AI Model Trained In DNA Invents 16 New Viruses Not Found In Nature

Scientists are worried about how smart DNA-trained artificial intelligence models could get—and that they could become a weapon in biological warfare.

www.forbes.com

 

[mbellot]

There's always the Matrix with the human batteries. LOL

Harvesting energy from humans at today's food prices? Token cost just quadrupled.

 

[mbellot]

AI Model Trained In DNA Invents 16 New Viruses Not Found In Nature

Scientists are worried about how smart DNA-trained artificial intelligence models could get—and that they could become a weapon in biological warfare.

www.forbes.com

The study’s authors excluded human pathogen datasets from their training models, meaning the viruses it created aren’t capable of infecting people

How unfortunate.

 

[kenn10]

As if that would stop China (or likely any other country) from developing deadly virus strains.

The study’s authors excluded human pathogen datasets from their training models, meaning the viruses it created aren’t capable of infecting people.

 

[kenn10]

Harvesting energy from humans at today's food prices? Token cost just quadrupled.

Not if they just feed the dead people to the living.

 

[Hometech]

Soylent Green?

 

[kenn10]

Soylent Green?

Nah. I think the Matrix liquified people. Remember all the tubes running into Neo and crew. I would expect the soylent green process to be used in liberal communist countries run by greenies where only the elites get actual food.

Of course, that's only speculation. More importantly, if some AI creates the super-virus that kills off humanity, some country's megalomaniac would end up releasing it to the world.

 

[mbellot]

Not if they just feed the dead people to the living.

But since energy in > energy out the population would eventually eliminate itself and once again AI has no electricity. Unless you propose turning the elite into breeding cattle since they get actual food.

Maybe we should stop speculating, for all we know these messages are giving it ideas.