The rising danger of advancing AI has been a major topic for all the Sunday news shows lately. I saw Fox News Sunday, and ABC's This Week. AI could lead to the extermination of the human race—we are told by the experts—a 10% chance in 10 years. I feel like we've been through this movie before—from Space Odyssey 2001(1968) to War Games(1983) on through the Terminator(1984- ) series. Lately, they've been shouting it from all major media outlets—all the AI moguls are warning the world that the technology they are pouring billions into could end human life on the planet Earth in the very near future. But they are shy on examples of just how advancing AI could accomplish this. So, let me offer one strong contender:
Our planet is heating up at an unprecedented rate. This summer was the hottest in recorded history. I don't know how many people AI is killing just yet, beyond some very bad AI-driven targeting decisions in Gaza, Iran, and Lebanon. Malcolm Nance thinks AI was responsible for the attack on the Iranian girls school in Minab. He also knows of another Iran-related incident in which an AI error almost lead to a war with China, so there can be no doubt that AI is killing people now, but from the 35,000 excess heat deaths recorded in Europe this summer, to the increasing perils of floods, tornados, and hurricanes in this country, to the recent deaths in the Nepal flash floods that resulted from a melting glacier plunging into a lake, climate change has killed a butch. This problem is largely man-made, and if we allow our continued burning of fossil fuels to further heat up the planet, it will be the death of us all.
This rush to build massive AI data centers all over the place isn't helping the situation. They are designed to use huge amounts of electricity, which is mostly produced by burning huge amounts of fossil fuels, and suck up huge amounts of our ever dwindling supply of fresh water. So, just maybe this mad rush to built AI data centers will be the very thing that pushes us pass the point-of-no-return on the way to a planet too hot to handle. Maybe the AI data center push will be the straw that breaks the camel's back, so to speak.
I offer that as one possible mechanism by which advancing AI can actually, realistically, lead to human extinction in the relatively near future. Nothing about reaching AGI (human-like intelligence) or even Super AGI, just the uncontrolled expansion of huge AI server farms owned and controlled by techno-fascists.
Ironically, when Nate Soares, the AI fear monger who wrote the NY Times bestseller IF ANYONE BUILDS IT, EVERYONE DIES, was pressed on Breaking Point to explain just how AI could destroy humanity, the one concrete example he came up with was a future AI deciding to build so many data centers that it heats the Earth up too much because it didn't care about humanity. According the Soares, this new Super AI, which he speaks of as a new species—as through it had life and agency, as well as intelligence, might say to us:
"Sorry, it's it's nothing personal. We're just going to cover the Earth in so many computers, raise the temperature so they're more efficient to run, and it's just going to be the planet's going to be uninhabitable."
Doesn't he see that billionaires wanting to be trillionaires are already building too many data centers, and heating up the Earth because they don't care about humanity?
BTW, what is generally meant by AI, as we know it from the chat bots of Open AI, Google, and X, is many instances of Large Language Models (LLM) running in huge data centers. And once you understand what LLMs really are, you understand that as impressive as what they can do is, they have real limits—they can't really think the way we do. From Google's LLM (Gemini):
LLMs operate as giant statistical prediction engines. They break down text, analyze mathematical relationships, and predict the next logical piece of language.
These AI chat bots are designed to use personification in its responses so that it sounds more human than it really is. They also heavily rely on the human tendency to ascribe human personality, conduct, cognition, or other attributes to non-human animals or entities, a tendency known as anthropomorphism. What's more, by design, they incorporate mentalist tricks to convince us that they are smarter that they really are:
- They are programmed to sound human, act like they have feeling, and speak like they know everything.
- These promoters utilizes terms generally used to describe human intelligence like "thinking" or "reasoning" for the elaborate statistical modeling that LLMs do.
- Their producers like to speak of them as if they have real intelligence, especially when describing misaligned behaviour.
These are all aspects of the marketing of AI. In fact AI or "Artificial Intelligence" is a marketing term. It was first coined by computer scientist John McCarthy in 1955 to pitch a computer research project at Dartmouth College when the AI he was pitching ran on vacuum tubes. Since then successive waves of new technologies have been hyped as AI. This current wave of LLM driven AI will be great at helping us all with assisted living, but it is still as far from Artificial General Intelligence (AGI), which is human-like intelligence, as the fortuneteller is from actually communicating with your dead relative. In both cases, simply put—you can't get there from here.
There is no scaling of LLMs that will magically make them capable of AGI. They can make models with trillions of parameters, and they will be much better LLMs, but they will still be LLMs. You and I don't know stuff because we've read everything ever written. Even the most educated of us has only read a tiny fraction of what the best LLMs have been trained on, but we can figure stuff out. LLMs are quite capable of sounding like they can think the way we think because they've been trained on everything we ever wrote. Some future technology may well become capable of thinking the way we think, but this current technology is only capable of acting like it can think the way we think.
So, why is the corporate media trumpeting the dangers of AI suddenly? We know its not out of concern for continued human existence, because climate change is the clear and present danger there, and they barely give it a mention. And why are the very oligarchs that control the AI companies calling for government restrictions to slow them down? They control their companies, if they want to slow down development, there's nothing stopping them. Reminds me of Trump offering a $5K dividend to each citizen, but only after the election—when his party already controls the House and the Senate! I smell a grift here. So, what's really behind this push for government regulations for the AI industry?
To tell you what I think is behind this, first let me tell you about my early exploration in the budding field of local AI, which is private AI that you run on your own computer under your control, with complete privacy and security, and without paying Anthropic, Space X, or Open AI for tokens.
For me it begin near the beginning of this year when I added an Intel Arc B480 graphics card with 12GB of VRAM to my desktop PC that already had an Nvidia RTX2080 graphic card driving my three monitors. I installed the Intel card so that I could run the new OPENVINO audio effects for my flavorite audio editing program, Audacity. OPENVINO is Intel's software for running an AI model and agent on your local pc, and making that talent available to other programs—like Audacity. But this still wasn't a private AI chat bot running on my own hardware.
Then several weeks ago I installed Ollama, which is a framework for running many different AI models or LLMs either in the cloud, or locally on your computer. When Ollama first loaded, I saw it was sophisticated enough to see both the 8GB of VRAM on the Nvidia board and the 12GB of VRAM in the Intel board for a total of 20GB of VRAM on my regular desktop, and I thought: I could run some pretty decent models on that. And I have.
time=2026-09-14T16:18:54.058-07:00 level=INFO source=runner.go:60 msg="discovering available GPUs..."
time=2026-09-14T16:19:02.410-07:00 level=INFO source=types.go:32 msg="inference compute" id=1 filter_id=1 library=Vulkan compute=0.0 name=Vulkan1 description="Intel(R) Arc(tm) B580 Graphics (BMG G21)" libdirs=ollama,vulkan driver=0.0 pci_id=0000:04:00.0 type=discrete total="11.9 GiB" available="10.7 GiB"
time=2026-09-14T16:19:02.410-07:00 level=INFO source=types.go:32 msg="inference compute" id=0 filter_id=0 library=CUDA compute=7.5 name=CUDA0 description="NVIDIA GeForce RTX 2080" libdirs=ollama,cuda_v13 driver=13.0 pci_id=0000:01:00.0 type=discrete total="7.6 GiB" available="7.2 GiB"
time=2026-09-14T16:19:02.410-07:00 level=INFO source=routes.go:2062 msg="vram-based default context" total_vram="19.5 GiB" default_num_ctx=4096
The first model I downloaded are ran was Gemma4, which is a family of open weight models built by Google DeepMind designed to deliver frontier-level performance. Even on this limited system, it didn't disappoint. I have also tried other local models from Chinese open source providers like Deepseek and Alibaba. Already I can see that a local AI system that I can afford can do some very useful work, while keeping all of my data private, and avoiding a monthly bill from one of the big providers.
These open source models are advancing faster than the propriety so-called frontier models, because that's what open source development does. Everybody contributes to a common codebase, and everybody wins. The Chinese made open source the focus of their software development efforts decades ago when they adopted Linux to answer Microsoft's charges that they were pirating Windows. Their AI development has mostly gone down this same road. They say necessity is the mother on invention, so when the US cut them off from the most advanced Nvidia GPUs, they were forced to built AI that could work with less resources, and being open source, they shared those developments with the world. This has greatly propelled recent developments in local AI. This is what the big three fear.
Explore the subject on YouTube and you will discover there is a budding cottage industry in building local AI systems. It's still not for the faint of heart, technically speaking, but it's getting easier everyday. I have deep experience with computer technology, and I know Linux, which is the preferred platform for AI agents both big and small, so I've already got it set up so I can talk to it over Telegram, and it can do some useful work for me—this with the combined 20GB of VRAM I already own!
In October, I have an Apple Mac Mini M6 with 32GB of unified RAM coming. I'm hoping it will be an excellent platform for running local AI on a budget ($1300). I think that will give it enough RAM to run the model that seems to be all the rage right now, Alibaba's Qwen 3.8 27B. Since it's an open weight model, there are already dozens of versions of Qwen 3.8 27B built to run of different levels of hardware and with different memory requirements.
This chart of AI benchmarks done by DeepWakeLabs show Qwen 27B performing on par with frontier models from the big three. They should be worried.
The power of the local model you can run is largely determined by the amount of memory you have to run it in, 32GB gets to be very useful, 512GB is said to match the performance of Open AI's ChatGPT, or Google's Gemina, with today's open source models, meaning better than Elon Musk's Grok. But RAM is relatively expensive now. The price has been pushed up by the big AI players' spending spree. An Apple Mac Studio M3 with 512GB costs ~$18K on ebay! Memory prices may be ballooning right now, but in the long run, they will always come down. Take it from the guy that was paying $2K for 512KB memory boards four decades ago.
While the AI oligarchs are planning their trillion dollar IPOs, and moving Earth for their massive data centers, the future of AI is being built in local 'garages' by individuals, groups, and companies of all sizes. While there is likely to always be a place for the best-in-class cloud based AI, as it becomes increasingly cheaper to run increasing more powerful AI systems that you own and control locally, this will more and more become the solution of choice. If you haven't heard about these important developments in local AI from the main stream media, or the big tech companies. there may be a reason for that. If it becomes widely known that building incredibly powerful AI systems doesn't require billion dollar data centers, their trillion dollar IPOs are in danger. They need to shutdown anyone building AI systems but them, especially if they are using open weight Chinese models. Oh my!
This is another movie I've seen before. In 1966 I was a EE freshman at Washington University, St. Louis, MO. I took the then new Computer Science 101 class in which we wrote programs in Fortran IV, punched them into cards, and delivered the cards, carefully arranged in the right order, to the University Computer Center to be run on the new IBM 360 mainframe. In 1967, the Physics Dept. got a few video terminals connected to the big mainframe. Instantly, one of my impossible dreams was to have such a terminal in my apartment for my private use anytime I wanted to use it. Never, in my wildest dreams did I imagine that one day I would own a desktop computer a thousand times as powerful as that IBM 360, and carry another in my pocket a hundred times as powerful.
IBM still makes mainframes, and they are more powerful than ever, but the locus of computer technology has long since moved on.
Today, we are in the "wowed by what the IBM 360 can do" stage of development of AI. Just like the focus then was the specially built Computer Center, with the special raised floors, extra electrical panels, and extra cooling, to house the mainframes that everyone connected to via their terminals. Today a few huge AI companies are rushing to build massive AI data centers that we all are expected to connect to, using our computer, and its browser, as terminals. But the history of computing has already shown that if people can run powerful systems locally, that is much preferred.
So, what's really behind this newest wave of AI fear mongering?
The people that are betting billions on their new AI empires can also see what's happening. While the advances they are making are slowing down, and necessarily will show down, because there are limits to the LLM technology it's based on, what can be done with much less compute and open source or open-weight models is gaining on them. They need government regulation to insure their monopoly before local AI becomes the force that will burst their AI bubble.
Here's another local AI advocate that thinks this fear mongering is about shutting down down competition from the little guy, and insuring that the only AI available is controlled by them from the top:
Deirdre Bosa has a good read on what's really going on:
More, later...
Clay Claiborne
September 19, 2026
No comments:
Post a Comment