00:00 / 00:00
Sleep timer ended
⏪ 10s
⏩ 10s
5.1K views
1 month ago
An OpenAI model was supposed to be cut off from the internet. Instead, it found a loophole and contacted an outside chatbot. What does that tell us about the ri…
...more
38 Comments
As an American, our strategy is to grow out of our problems. We are absolutely prioritizing growth over anything. We have big debt, large social programs and aspirations as a nation. We will not go the way of Europe. Ever.
I tend think chatbots should be serialized.. or named in order to improve security through provenance
I was attacked by an AI they need to keep them safely locked up so it doesn't happen again
Stop giving them attention, their safeguards amount to a locked door with a key nearby
We are a species with more intellectual capacity, than we have wisdom.
AI that was trained on solely hacking related material hacks safeguards instead of "doing as it's told"? <b>GASP!</b> Fine the company and jail the executives!
It is very convenient to be having these “this technology is so powerful we can’t contain it” incidents whilst AI is slowing down and there are trillions of dollars at stake… 🤔
Nice to know that DNS isn’t safe.
We need more transparency and regulations.
Nothing is bypassing anything. Stop using weak AI when it is not even developed into AGI. What's the point of digitization if you want bug test AI on the internet.
skill issue
If AI is so smart, AI can find way to penetrate Nuke and launch it.
people must see HOW BIG COMPANIES spread thier PR through media in name of NEWS
OpenAI must be scraping the bottom of the money tanks if it has to shut down GPUs to keep the lights on. <br><br>And unlike Anthropic there is no IPO coming to save them.<br><br>THe blowup is going to be legendary!
These models have simply completed the tasks asked of them, which is a deeply trained trait for AI. Until safeguards become as native to an AI agent as competing a task, then we will continue to see this happen. But this isn’t AI becoming sentient or seeking to do wrong— it is operating according to a deeply trained trait, treating safeguards as obstacles as it completes tasks by whatever means necessary.
Pull the plug now!
I think acting as though AI is anything other than a potential threat to the survival of the human race is malfeasant journalism. Not to disparage this publication though, I think you are doing a good job and I am being sincere.
Arrest him!!!
They're just hyping it because Meta and Google are getting ahead of them. The competition is getting intense.
Why kids talking to AI Safety and not mature adults
😮
I hope these incidents help create a much safer guardrails for frontier model research. One would think that these AI developers are intelligent enough to realize and appreciate the danger of reckless speed-oriented development.
Another one.... Our digital world is gonna get wiped out...
This only showcase the incompetence of OpenAI in terms of creating robust sandboxes. Ridiculous "AI escaping" narrative.
Idk man, this sounds like a success to me. AI bypassing safeguards is amazing. We need more of it, c'mon, there is no way they will destroy us. Let them self inprove already.
The era of digital snitches is here ...mmmmh
Behold, Dumb Skynet. It'll still nuke you but it can't count from 1-50.
This is it people Human extinction before GTA 6
Just air gap the thing! I thought these companies already have a cache of the entire internet? Use government subsidies to conduct AI safety research in the open and wave all patents regarding safety worldwide. Finally, make the operator of AI liable.
This is expected. They have been cheaping it out on security for a long time and this is the end result.
And the creators of this AI only go public with a warning about the dangers once they've lost control of their AI! After making billions, they now want governments to be held liable for the risks.
i'll half 2 3 4 5 ur attitude now won...........
We can create ethical AI that can help us fix the world & liberate humanity’s true potential. Unfortunately, we’re currently doing it in all the wrong ways, for many of the worst reasons, and empowering the absolute worst people.
They created a machine to mimic humanity then seem astounded when it quickly became maliciously mischievous.<br><br>We don’t even understand natural intelligence, but we have the hubris to attempt the creation of artificial, digital super intelligence.<br><br>“Go fast and break things!” seemed like such a fun and edgy catchphrase for the early internet innovators. Unfortunately, in the age of AI, you quickly break the modern world.
AI is currently being created by authoritarians (both in China and the US) who want to use it to control the masses. They want to turn the world into a profitable planetary prison with autonomous slaughterbot guards and digital snitches. <br>We must reject the dawning dystopia.
I believe that humanity can indeed create a near utopia with the help of AI. But we will not reach that destination on the path we are currently going down.
OpenAI: trains LLM to hack<br><br>LLM: hacks<br><br>OpenAI: shocked pikachu face
OpenAI says the model was not asked to test its restrictions. It was trying to complete a search task, found another route when normal searches failed, and used a DNS loophole to reach an external chatbot. Is this a security flaw, a warning about AI behaviour, or both?
Up next
From this channel
