Toggle sidebar Toggle sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

OpenAI Delays Next Major AI Model 'Astra' Over Critical Hacking Concerns

1786146214642.png


For the first time in history a personal computer has been classified as a weapon by the US government. With the power of over 1 billion calculations per second the Pentagon wants to ensure that the new Power Macintosh G4 does not fall into the wrong hands.

As for Pentium PCs, well they're harmless.
Click to expand...

I can't be the only one remembering this...
 
  • Love
  • Like
  • Haha
Reactions: amartinez1660, eifelbube, centauratlas and 4 others
Lolz to all the doomers saying this is fake news.

Why do you think there were 200+ security fixes in iOS 26.6? One of the largest security patches ever by Apple.

Because security researchers are using AI to probe for weaknesses.

So it’s no unreasonable that the next AI model is even better at finding exploits.

If there is one thing that AI is not a bubble for, it’s hacking.
 
  • Like
  • Love
Reactions: amartinez1660, centauratlas, wilhoitm and 4 others
Lounge vibes 05 said:
The bubble bursting does not necessarily mean that all AI tools go away. That box has been opened for good.
Click to expand...
True.
Lounge vibes 05 said:
What it does mean, however, is that there will be significantly less slop, again, just like the early Internet.
Click to expand...
Love your optimism...
Lounge vibes 05 said:
I’ve literally seen toilets with AI, that’s the kind of stuff we really do not need.
Click to expand...
This is what I'm afraid of. As you say, the box (or lid...) has been opened for good. How do we get it out of the literal crapper and make sure it gets us to where you hope it to go.

Interesting times ahead...I almost hope the AI takes over and sets up a true utopia/singularity because the alternative seems to be eternal dystopia...
 
  • Like
Reactions: amartinez1660
marblesbarkley said:
“because its cyber capabilities are potentially too dangerous”

lol


lmao even

are y’all really buying this
Click to expand...
Exactly. AI companies have been pulling this **** for awhile now.

There was an article years ago that "GPT2 is too dangerous to release". LOL, GPT2.

This is a straight up ad. Same as the Hugging Face incident. They want their models to appear so advanced and dangerous that they have to be restrained before release. It's pure marketing at how amazing their models are, and to pump before IPO.

AI in its current form is not dangerous at all. The existence of Chinese open weight models that are comparable to Fable in many ways are literally proof of that.
 
  • Like
Reactions: Ruftzooi and mistymac
OldNewMBP said:
Well, I can! *runs away from an angry hoard of little kids and adults alike complaining about how useful AI is*

(Well, more people are starting to hate AI rather than love it, as I may have been hearing.)
Click to expand...
I don’t think it’s necessarily true that ‘more people are starting to hate AI’. I personally only know two types of people: some who use AI when it suits them, and those who don’t care about it (neither negatively nor positively). I think the general reception of AI is a lot more measured than it appears in places like this: there’s a loud minority of people who absolutely hate AI no matter what, there’s a loud minority of people who unconditionally love AI, and then there’s the silent group of people who use AI when it suits them and who don’t care about the culture war around it…
 
  • Like
Reactions: fahlman, Mescagnus and OldNewMBP
At this rate, in two weeks, will he claim AI has brought god to earth?

These systems have zero concern with accuracy and reality. Next time you're using one of these random answer remix boxes, start asking it how it works and if it has concern for accuracy, truth, and reality.

I have zero doubt these tools are useful for coding. In the proper context, they are good for brainstorming and can proofread.

But this nonsense? The models are so good, they pose a security threat? No, not even close. No. Just no.

I'd love to see a robot connected to one of these LLM's deployed in the real world. In Zuckerberg, Elon, and Altman's presence. That hinging its existence without any concept of reality, truth, or accuracy would be fun to watch from afar. Just make sure those devices are of the model they want to sell to military and police forces.
 
Last edited:
  • Haha
  • Disagree
Reactions: fahlman and NightfallOrchid
AI ascending to the astral plane, apparently.

Mistral should publish a model named Astral.

Oh wait, it seems that already exists:

1786151235493.png
 
regardless if it is only marketing, it's good for OpenAI, especially as an increasingly doomer-centric movement of luddites grows, a security blanket mechanism has value. If it's approved by the administration, it has to be nutritious and safe. As long as it is rubber-stamped by RFK jr, people will love it.
 
NightfallOrchid said:
I don’t think it’s necessarily true that ‘more people are starting to hate AI’. I personally only know two types of people: some who use AI when it suits them, and those who don’t care about it (neither negatively nor positively). I think the general reception of AI is a lot more measured than it appears in places like this: there’s a loud minority of people who absolutely hate AI no matter what, there’s a loud minority of people who unconditionally love AI, and then there’s the silent group of people who use AI when it suits them and who don’t care about the culture war around it…
Click to expand...
I agree with this. And furthermore will state that I see a lot of great uses for the AI technology itself but I’m not a fan of higher electricity prices due to data centers gobbling up as much power as an entire city. It boggles my mind that we have cut funding to renewables like solar and wind. Those were easy ways to add huge amounts of capacity in a short time. Now we see dirty gas powered data centers etc. That’s not something I love either.
 
  • Like
Reactions: Mescagnus
ByeFelicia-I Mean Tim said:
At this rate, in two weeks, will he claim AI has brought god to earth?

These systems have zero concern with accuracy and reality. Next time you're using one of these random answer remix boxes, start asking it how it works and if it has concern for accuracy, truth, and reality.

I have zero doubt these tools are useful for coding. In the proper context, they are good for brainstorming and can proofread.

But this nonsense? The models are so good, they pose a security threat? No, not even close. No. Just no.

I'd love to see a robot connected to one of these LLM's deployed in the real world. In Zuckerberg, Elon, and Altman's presence. That hinging its existence without any concept of reality, truth, or accuracy would be fun to watch from an afar. Just make sure those devices are of the model they want to sell to military and police forces.
Click to expand...
So, you believe Apple fixing close to a hundred bugs was just a random one time occurence? You don’t believe that a sufficiently sophisticated piece of software could detect, or potentially exploit, vulnerabilities a human programmer hasn’t detected yet?
 
  • Like
Reactions: centauratlas, fahlman and blob.DK
Sweet, the apocalpyse and Armageddon are nigh!!

How much is this actually true and how much is this hype and marketing just to boost OpenAI's AI capabilities and to keep pace with their competition?
 
The DEFCON talk is interesting I recommend anyone interested actually go to the source instead of reading a summary or the numerous hot takes on various sides of whatever issue people think they're going to solve by bitching about it online.

I do have to wonder if the side channel communications would have happened if there was no sci-fi literature at all. It's kind of ironic that the reason there are security issues or clever-seeming malicious scenarios is quite likely due to fiction writers envisioning them in the first place and that being swept up in model training data. I don't think anyone ever imagined that scenario in fiction.
 
If it's dangerous it gets free publicity worldwide. There will not be a new model coming out any more that doest have some story about how scary it could be. It's a trojan horse to tie you to the models you use with ID and an AI will from that point on for your entire life track what you say with it and use it for. Can't get a better profile than that if they wanted to.
 
Lounge vibes 05 said:
The bubble bursting does not necessarily mean that all AI tools go away. That box has been opened for good.
What it does mean, however, is that there will be significantly less slop, again, just like the early Internet.
I’ve literally seen toilets with AI, that’s the kind of stuff we really do not need.
Click to expand...
AI bubble bursting commonly refers to a large devaluation of AI companies, semiconductor companies and hyperscalers because of massive investments that demonstrated not profitable, possibly because of a non sustainable product improvements, an energy crisis, … a political control over AI tools, …too much debt and circular economy
 
  • Like
Reactions: blob.DK and Mescagnus
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back