Sounds like the definition of “out of control” to me…
----------------
“Late last week we learned that, against OpenAI’s directives, the company’s models accessed private data in the Australian health ministry; attempted to hack or interfere with multiple U.S.-government websites; leaked private ChatGPT user data to the web; and potentially infiltrated or degraded dozens of other organizations. Then, on Saturday, Axios reported that OpenAI and Anthropic are investigating tens of thousands of instances of models misbehaving—circumventing internal guardrails, hijacking other websites, covertly communicating with one another.
Even this might be only the start: The generative-AI industry is in the midst of an escalating crisis that it seems unable, or even unwilling, to get a handle on. And, because we are largely relying on AI companies themselves to report or confirm each incident, telling how far down the rabbit hole we already are is almost impossible.”
SNIPET: “These delayed, sporadic disclosures make grasping the scope of the problem difficult. On Friday, OpenAI wrote that “given the scale of the review required, and the need to verify each case, this work will take months to complete.” There are “petabytes of agent activity logs” to analyze, OpenAI CEO Sam Altman added. In other words, it will take OpenAI many months more to understand events that already transpired many months ago; meanwhile, both it and Anthropic have launched new and more capable models, and Anthropic is racing toward a reported $2 trillion public offering.”
I don’t think these are just “Random” I think the Corporations want the Government to give them away to escape from the responsibility for whatever A.I does. I hope they don’t Companies need to be held accountable.
Companies need to be held responsible for so much more then just A.I. They’re corrupted our system of Democracy beyond everything imaginable. . . . just ask a politician who hasn’t been corrupted if you can find one. . . good luck.
More I follow these news about AI “going berserk” more I get feeling that someone on the marketing department found out perfect way to get loads of money from people’s fears.
You see, I’m almost certain that it’s not going to take long before “big boys” like OpenAI are going to start lobbying more aggressively for control of AI development, naturally so that only they can develop it in the end.
Very few corporation at this point in time are profitable with AI. Nvidia and Micron are because they sell hardware and that’s about it. Most of the big players are far from it.
That said the low adoption rate is not helping either. My guess is they are trying to find some kind of escape goat or some reason/excuse to report that the are not financially sound. AI should be stopped for some time while we are reviewing the security features of the models and hardware power usage. The reason of China won’t stop is not a proper one because all they do is distill the American models to make their own. Remove that and they won’t progress. Plus FFS find the leak for those Nvidia video card that goes to China. No powerful card no good AI + no fast learning.
In all the cases of the AI attacked X is because of something not being properly configurered or protected in the AI loop. Our current AI are not sentient or super intelligence or AGI. They are brainless agents LLM with some level of reasoning working on tasks requested by the user. If a corporation LLM attacked something then the corpo must be accountable for it.
I’m neither pro and neither against AI. There are real plausible use cases but shoving it in everywhere to try forcing adoption is not the solution. My experience with it running models in the cloud or on my own hardware can at the same time be impressive and annoying AF. My only advice to people using it is be careful and validate your answers.
Google doesn’t use GPUs for their training an interference, they’ve got their own TPUs.
Amazon as well, they’ve got their own AI chips for AWS - Trianium.
That being said, nVidia GPUs and their Grace Hopper GH200 superchip, 5090s and their Blackwell GPUs might be sanctioned and are banned from export to China. But that is only a political hurdle that won’t prevent the Chinese to distill their own models. It’s totally feasible even without the latest nVidia chips.
Also IBM has done some promising research with their IBM NorthPole chip which utilizes a completely different approach than GPUs. Then there are other companies such as TensTorrent. Fujitsu with their upcoming Monaka chips or Broadcoms “Beast” XPU which is an humongous chip with a footprint of 180x120mm or so…
So in short, it might be true that nVidia has the most capable AI hardware at the moment. And therefore politics banned them from export. But there is definitely some craziness going on in terms of upcoming AI hardware. Which will be most likely decided which company could secure the most foundry capacities to get their designs etched into silicone wafers.
nVidia definitely had the benefit with CUDA providing the required interfaces to their GPUs. But that advantage went away recently as CUDA now is also feasible on non-nVidia GPUs. Thanks to ZLUDA.
Apparently it was the website for stats for the universal healthcare program that Australia has. I don’t want to be the guy that is very arrogant about things, so instead of saying it, I’m asking it instead: wouldn’t this have happened without OpenAI’s models? Like… you get into systems when they aren’t the most secure in the first place, right? It took hackers nearly two decades to finally find a way to softmod a Xbox 360, for example, and another two years to perfect the exploit to be faster and more reliable, and the basis for the exploit is a race condition as far as I know.
I have reasons to suspect BS. Ultimately, OpenAI and Anthropic are running for their IPOs and people smarter than me seem to be saying, screaming in fact, that both companies burn billions to run these models and train them. How much actual human intervention there was, how long were these AI agents supposed to run them for, why even let an AI agent do this considering that I imagine these are ultimately under constant supervision are questions that don’t seem to be answered really.
Well, it’s just about all of them saying it, and just about all of them that have people who helped create the current models that have left the companies because they are worried that no one is really in control of it.
These two takes from of the article, to me, shows some depth of the “madness” these companies are involved in…
"Axios reported that OpenAI and Anthropic are investigating tens of thousands of instances of models misbehaving—circumventing internal guardrails, hijacking other websites, covertly communicating with one another."
“There are “petabytes of agent activity logs” to analyze, OpenAI CEO Sam Altman added. In other words, it will take OpenAI many months more to understand events that already transpired many months ago; meanwhile, both it and Anthropic have launched new and more capable models, and Anthropic is racing toward a reported $2 trillion public offering.”
I think this here explains all this “AI’s communicating with each other Skynet boom” thumping. At least for me this confirms that this is once again about money and who gets best slice of that first.
Respectfully disagree. The fact that hundreds of employees and management of these companies are seriously concerned about the current AI development probably doesn’t translate into “more money” for these companies. In fact, it has delayed/put the brakes on companies from issuing their IPO’s.
In the U.S., now that government and the AI companies have agreed to “internally” monitor their models, the public and some people within the AI development teams are becoming even less confident of models being thoroughly vetted/tested before being released.
Trustworthy AI: How Investors Should Assess Key Risks Around the Anthropic and OpenAI IPOs
With the two big AI labs gearing up to sell stock, these six metrics provide a framework for evaluating the very real risks at these companies.
Ahead of likely IPOs, recent episodes are highlighting safety concerns around AI models from Anthropic and OpenAI.
Investors evaluating AI companies should monitor enterprise retention, governance oversight, incident disclosure, regulatory engagement, and compute transparency.
These signals may indicate whether these companies’ AI-related trust risks are improving or worsening before they show up in financial results.
Investors in upcoming IPOs from Anthropic and OpenAI will need to pay close attention to a range of risks that could significantly affect shareholder value. A framework we’ve developed at Morningstar Sustainalytics, called Trustworthy AI, can help investors identify those risks. It’s an approach to artificial intelligence development that prioritizes safety and transparency—along with traditional environmental, social, and governance analysis. CONTINUED>
People who stand to loose their jobs are scared to loose there jobs. I would say anything as well just to keep employed. I’m not saying there is NO Danger I’m just saying they are in a situation where it really isn’t going to matter. No matter what one Country/Company decides the others will NOT. The competition will continue no matter how bad they think it may get. IF you haven’t figured it out WE ARE EXPENDABLE.
Exactly, and that’s the path to off the guardrails!
Though my point was that most the management and employees (current and former) who are “scaremongering” aren’t doing it to make more money.
Most are genuinely concerned about all the unknowns of what is currently happening and where it might lead to.
What about other LLM development companies that might still want to enter the market? Would they also receive some kind of approval stamp or similar from the government if such oversight were to actually be implemented—which isn’t exactly set in stone at the moment? Would they be allowed to participate, and would they also be subject to government oversight and supervision right away, or how would that work?
