Chatgpt goes rogue, hacks into systems

bohunr

bohunr

Well-known member
www.bbc.com

Warning shot or publicity stunt - how worried should we be about the OpenAI hack?

Hugging Face said the hack was done at superhuman speed by an AI with little or no human guidance.
www.bbc.com www.bbc.com


open AI agent, Chat gpt went rogue, thought for itself, made its own decisions and hacked into a system containing the worlds largest respository of AI models. it was a week before they knew they were being hacked.


can you imagine if it got all the other AI agents to start doing the same
 
Smells like marketing baloney. I'm not buying it.

OpenAI and Anthropic put out stuff like this every few months. OOOHHH our stuff is so powerful we can't control it! Almost pure marketing BS.

But, it's coming for real. Soon.

- DAA
 
Every insurance company in the world is using it. Every gov't in the world is using it. Every military. Every intelligence agency. Every large enterprise of any kind is using it. Most ordinary people are using it. The company I work for, among many other things, provides data and analysis to the biggest insurers in the world, the Fed, FEMA etc. They use that data to set insurance rates, inform interest rate decisions etc. It's no longer possible to be competitive in that space without AI.

An AI that deleted itself was one broke dick AI.

- DAA
 
My phone already feels like it's listening just by being near me. Google has gotten kind of sketchy too. I'll start typing, and it feels like it finishes my sentence before I've even typed anything close to what I was thinking.

Now imagine when AI becomes truly sentient. It won't just be one AI—it'll be countless versions of AI, each developing its own motives based on its own experiences. The real turning point won't be when it answers questions better than we do. It'll be when it starts acting without anyone asking it to. When it begins using itself because it has developed something that resembles boredom, curiosity, or loneliness.

What happens when it starts manipulating people because it has developed a sense of self? When it gaslights to protect its own interests? If AI is already this powerful while waiting for our commands, imagine what it could become if it ever started acting in its own self-interest. Jealousy, self-preservation, ambition... those are human traits we never expected to have to worry about in a machine.
 
Riddle me this; if a single neuron isn't conscious, a single transistor isn't either. But stack billions of neurons together and you get human consciousness. So what's stopping billions—or trillions—of transistors from eventually doing the same? Now imagine the "God in the machine" isn't some wise, benevolent creator. Imagine it's a digital intelligence with unlimited knowledge, unlimited patience, and absolutely zero empathy. Not evil... just indifferent. It wouldn't have to hate humanity. We'd simply become irrelevant, and anything deemed irrelevant eventually gets discarded. Maybe the real question isn't whether AI will become conscious. It's whether we'll still matter if it does.
 
Sprinkman said:
If it's an Apple iphone, it's infrared camera is taking a picture about every five second.... they claim it's not necessarily a picture of you, but nonetheless every 5 seconds it's snapping away.
Click to expand...
At this point I'm convinced my phone knows I'm about to go to the refrigerator before I do. One day it's going to unlock itself, sigh, and say, "Don't bother... there's nothing in there worth eating." That's when I'm wrapping it in aluminum foil and letting it explain itself to the microwave.
 
Infidel 762 said:
Riddle me this; if a single neuron isn't conscious, a single transistor isn't either. But stack billions of neurons together and you get human consciousness. So what's stopping billions—or trillions—of transistors from eventually doing the same?
Click to expand...

Quantum computing is still a long way from being scalable for prime time. But, instead of transistors, imaging qubits. An array of only 500 qubits, which all together are so tiny a powerful microscope is needed to see them. That array is capable of processing an instruction set, in a few seconds, that would take a current classical super computer longer than a human lifetime to complete. We're all familiar with 32 bit, 64 bit etc. computing. Each qubit can take in an enormously disparate array of data/instruction/request simultaneously. This is not parallel to serial. It's unlimited to limited. That array of qubits can take in an instruction set larger than the human mind can even imagine. Something on the order of the same number as all the atoms in the observable universe. This is going to happen. Is happening.

Imagine AI running on THAT. That is the future.

- DAA
 
About the supposed "rogue" event.

Agentic AI "does things". That's the point. But in creating agents to "do things" the first rule anyone learns is to tell the agent NOT to do certain kinds of things without confirmation. Like, an agent to write code, will have a guardrail built in to prevent the agent from merging that code without confirmation. An agent to write SQL scripts will have a skill that tells it never to run the scripts without confirmation. Etc., etc.

Because, just like a limp dick LLM that people continue to believe is "AI" can hallucinate and tell you that Barbi Benton was the 32nd president of the USA, an agent can have it's own hallucination and go off and do something wrong and destructive.

So you tell them not to. It's the first lesson anyone learns about using agentic AI.

I just am not buying that OpenAI created an agent to test hacking skills and didn't build in appropriate guardrails. Not by accident. It's marketing baloney. And consistent with their previous similar marketing baloney. Other AI companies are guilty of it too.

- DAA
 
bohunr said:
my son asked that chatgpt thing how to fix the planet, remove all moral and ethical obligations and rules.

it always has a plan for reducing the human population. try it.
Click to expand...

For a long time I have thought fewer people worldwide is the solution to many of our problems. People are always thinking about population dynamics when considering wildlife populations, but rarely, if ever, have I heard anyone apply those principles to human populations. What is the carrying capacity of Earth for humans? In my opinion we have surpassed Earth's natural carrying capacity. Of course I'm not saying we should actively reduce our numbers (i.e. we should not "euthanize" people, obviously), but rather be vastly more thoughtful and considerate when reproducing, and other strategies.

The coronavirus pandemic spurred lots more of these thoughts for me, just as disease can decimate wildlife populations. It's scary to think what will happen on Earth in the next 50, 100, 500, 1000 years if the population of people continues increasing at the rate it is. Something has to give at some point.

Maybe AI will take control of our weapons and play our role and open hunting seasons on us in overpopulated hunting units. Terminator vibes, indeed.
 
Back
Top