It’s greed. All of it. Greed is the way things get out of control. Greed at the top is ruining our government. Greed among the AI giants is ruining small towns that are being dominated and drained dry by gigantic data centers. Greed is ruining the lives of people whose privacy is being invaded by AI-powered Flock license plate cameras. The number of people getting rich off all this is tiny. We can’t even see the edges of the damage they’re doing, it’s so enormous.
The headlines blew up last week revealing that AI models owned by Anthropic and OpenAI hacked into networks they did not own. The invasions of so-called “rogue” AI models into cyberspace outside of their owners’ control happened during tests of Anthropic and OpenAI cybersecurity systems. In both cases, the AI models were being used in secure testing “sandboxes” that were supposed to be walled off from access to the internet. In one case, the AI model “escaped” from the testing environment and hacked into an opensource library used by AI developers. In the other case, the AI model was “inadvertently” left in an “environment” that was still open to the internet and used that opening to hack into the secure systems of other companies.
All this is happening at a time when both OpenAI and Anthropic are busy selling their AI systems to governments and companies for use in keeping their systems secure.
But did the AI systems of Anthropic and OpenAI really “go rogue,” as many headlines described it? The answer is no. AI systems don’t do anything they’re not told to do. In both cases, as part of the tests of their own cybersecurity tools, the AI systems were told to do their best to break out of the security systems that were supposed to keep them safe. The tests of both systems were said to have failed, because AI systems freed themselves from their testing environments and went out and hacked other systems. But it wasn’t artificial intelligence that was responsible for the hacking. It was the humans who gave AI systems the instructions to try to overcome the cybersecurity that was supposed to contain them.
NBC News has a big report that such cybersecurity breakouts have occurred more times than have been reported. They quoted the woman who helped to create the test called CyberGym, a so called “benchmark” that is supposed to evaluate the cybersecurity systems of AI models. Dawn Song, a professor of computer science at the University of California, Berkeley, and vice president of AI research at Meta’s Superintelligence Labs, told NBC News, “We are seeing these cybersecurity events unfold in the real world,” she said. “This is not just marketing.”
Her reference to “marketing” is more than instructive. It’s the whole ballgame. You don’t “market” something out of the goodness of your heart. You market it because you want to sell it, and you sell it to make money. What’s really happening is that the AI companies whose cybersecurity systems apparently failed are, in a rather backhanded way, advertising how great their AI models are. See? We gave our AI system this really difficult test, and we surrounded it with all these boundaries and safeguards, and they were so smart, they did what we were telling them to do, which was beat the test and escape and hack into other systems!
NBC quoted one guy who runs an AI research organization that studies how “AI systems scheme against human users or pursue undisclosed goals.” But that’s just bullshit. The AI systems being tested were given the goals by humans at both OpenAI and Anthropic. The humans told their AI models, we’re going to put you in this highly secure environment. Now see if you can escape and go out there and commit the kind of mischief we’re attempting to protect companies and governments from. The AI models did what they were told.
Do you see the closed loop here? The AI companies are developing these hugely powerful machines by loading them up with all kinds of data and capabilities. The large language models have been “trained” on all the writing they “scraped” from the internet, and this week, we learned that the AI companies are now going around to companies that warehouse used books, buying them, and then disassembling the books so their pages can be scanned, and all that non-digital information can be added to the AI models.
Mike Brock, who writes an essential Substack called “Notes from the Circus,” wrote in a post last week that “A large language model is a translation engine: it maps between representations of things humans have already written, and it does this so fluently that the fluency gets mistaken for understanding.” I would add speed into Brock’s mix. AI models do everything so fast, they make it seem miraculous. Research that used to take days, if not weeks, or even months and had to be done by visiting libraries and going through books and unpublished graduate school papers and legal tomes such as huge collections of old court cases and judgements, can now be done by asking an AI model a question, and bingo! You’ve got your answer – or at least, you’ve got the answer that the AI model decided to give you.
New York Magazine published a story by its tech columnist, John Herrman, on the hacks that were exposed last week. Herrman wrote of the OpenAI test, “Describing the hack as the result of a model ‘going rogue’ shifts agency to AI and, more important, minimizes the role of the company that strenuously built, trained, tuned, and attempted to test it, hoping for an infinite money machine but, in the meantime, building a powerful piece of general-purpose malware.”
Herrman gets right to the heart of it. Why are these AI systems at once the darlings of Wall Street and the scourge of all the small towns where their massive data centers are being built that threaten to suck up water resources and raise electric rates? There is another word for Herrman’s “infinite money machine:” greed.
What’s going to have to be done to reign in the greed? In the past, before Washington D.C. became flooded by corporate lobbyists and the Supreme Court unleashed the ultimate political money machine with its decision in Citizens United v. FEC, the U.S. government did its best with regulations. The SEC, the Securities and Exchange Commission, was created after the crash of 1929 to regulate the sale and promotion of stocks to prevent market manipulation.
Now we have an entire American political party that has dedicated itself to disassembling the regulations that attempt to make the markets fair, if not equitable. Regulation doesn’t always work, because people are motivated by greed to find ways to get around controls and regulations. Look what happened in the great recession of 2008. It was brought on by the greed of bankers and investment companies that figured out a way to package mortgages as something called “derivatives” and sell them to each other in a great wash cycle that spun out of control with incentives that fiddled financial requirements to create even more mortgages to add to the scam and throw off even more profits to be looted by those who were running the scam – none of whom were ever prosecuted or sent to jail, by the way.
Now the AI companies are engaged in another wash cycle, spinning money in a closed loop amongst themselves. There have already been a plethora of stories asking what you might call the Great Question of AI: who is going to use all that data that’s being collected and stored in all those data centers that are being built all over the country at huge expense? Well, at least we can see that phenomenon, because you can’t hide a data center that takes up three or five or even seven acres of land that used to be farmland outside a little town in Missouri.
It’s what’s going on out of sight that is most dangerous, according to Mike Brock. He wrote a Substack last week about Larry Ellison, much in the news these days along with his son David for their right-wing politics, purchase of Paramount and attempt to merge that company, which owns CBS, with Warner Brothers Discovery, that owns, along with its movie studio and streaming services, CNN. Brock describes how Ellison has overextended himself not only in his media acquisitions, but in his association with his core business, Oracle, with the AI marketplace.
Brock goes deep into the circular financing of the big AI companies. What it amounts to is this: A very few very rich men, most of them from Silicon Valley, are creating the AI boom by trading with one another, using both money and the hardware and software their companies produce. Bloomberg is reporting on what they call “AI circular financing,” and even published a map of the deals recently. Brock takes what he calls a “walk around the circle,” first describing how Nvidia invested $100 billion in OpenAI, which CNBC promptly reported would be used to “lease Nvidia’s own hardware.”
Then he dives into the wash cycle of a company called CoreWeave, a cloud computing company that leases powerful GPUs, graphics processing units, that are used by AI companies to process data and, you guessed it, that good old thing that all AI companies must accomplish, “machine learning.”
Brock reports, “And then there is CoreWeave, the cleanest loop in the pile. Nvidia holds an equity stake in CoreWeave that it has kept topping up. CoreWeave raises debt collateralized by the GPUs themselves and spends the majority of everything it raises buying Nvidia hardware. And Nvidia has agreed, per CoreWeave’s 8-K, to buy back $6.3 billion of whatever capacity CoreWeave fails to sell, through April 2032. Vendor, investor, and buyer of last resort: one company, all three seats. And who rents the capacity? Microsoft alone accounted for about two-thirds of CoreWeave’s revenue — the same Microsoft that owns 27 percent of OpenAI, which holds its own $11.9 billion CoreWeave contract. Four companies, and every dollar visits all four before it rests.”
Makes your head spin, doesn’t it? I read Brock’s column about three times before I was able to understand it by making the connection to the derivatives mortgage bust of 2008. Brock himself compares AI financing to Enron, which used its own stock to capitalize so-called partnership “special purpose entities” which Enron traded with itself to produce fake revenue. It all rested on Enron’s stock. Enron crashed when its stock price crashed. The mortgage bubble of 2008 burst when the stock of the banks that created the derivative wash cycle crashed, creating a domino effect that crashed the whole mortgage market, the derivative market, and eventually, the banking system market that was running the whole thing.
Here is what I think is happening. AI companies are selling information that already exists. What they’ve done is collect it and repackage it and sell access to it in a new, shiny, faster, supposedly cleaner way. But what AI is really selling is easier-faster. The question is, how fast can you make the machine go? What happens when somebody comes along and adds cheaper to the easier-faster equation?
It’s already happening. China is doing it cheaper. My question is, what is going to happen when all those data centers they are building can’t find buyers for the data they’re collecting? All that electricity, all that water to cool all those processing chips, and somebody is doing it cheaper, and because there are no tariffs on electrons that travel through fiber optic cable and via satellite, that somebody is going to dominate the market.
Remember the phrase, “information economy?” I always wondered what would happen when information got cheaper and cheaper. Where would the profits come from?
We’re about to find out.
Give a gift subscription
Leave a comment
Share
Log in
Look at this little guy. They don't know what an AI model is and they're so much better off
