More Challenges Facing AI

Is it a bubble?

There’s been a lot of news about AI recently. Goldman Sachs published a paper this week saying while $1 trillion is scheduled to be invested in AI capex over the next few years, there is no sight of the revenues generated to justify such expense. As I have previously mentioned, Sam Altman has even expressed shock that AI has not had a noticeable effect on the economy yet.

While promises of AI replacing workers and lowering costs while improving life abound, are they real? Well, I don’t know but I’m glad to give my opinion. To do so, I did a bunch of research this week into LLMs, what OpenAI and most other so called AIs use at their core. See, LLMs take in a huge amount of data (most of the internet) and train a model that is really only good at predicting, say, the next word in a sentence. To do this, it just uses a a ridiculous amount of fairly basic linear algebra. While the concept has been around almost a decade, it’s taken a while to refine it and have the immense processing power be available. Training these models takes months and reportedly up to a billion dollars of compute power and then, they don’t learn anymore. They are stuck in time. Try it out on ChatGPT for instance. Their latest version was trained late last year and knows nothing about anything since then. AIs are stuck in time and not only cannot learn anything new while updating them takes the aforementioned months and billions in compute power.

That is to say, they don’t actually do much other than produce text, voice, and images. It is knowledgeable, in that if you ask it a question it’ll most likely come back with the right answer, but it’s not intelligent and it’s also, more importantly, incapable of doing anything else. You can’t ask it to book plane tickets or order your groceries. All it does, for the most part, is spit text at you. iOS 18 with it’s new AI features will finally have some sort of integration with apps on your phone but that remains to be seen how useful it’ll become. It’s not until AI crosses into the real world that true value can be created. There is only so much to be gained by something that just generates text. To summarize a great plot on South Park, AI can only replace jobs that don’t require opposable thumbs.

So, do I see AGI or even AI as a big threat? No. It’s today’s fad like self driving cars were around 2016-2017. Back then, all you ever heard about was how every car would drive itself by the early 2020s. Hundreds of billions were invested. And what came from it? So far huge capital losses and a few small working prototypes like Waymo. I see AI as very similar. It’s a lot of capital chasing a dream without knowing where the money is coming from not to mention its technology that can only go so far. While I do think so called AI is here to stay since it can be very useful, I think the hype will die down and we’ll see more of a slow growth of its use unless some unique breakthroughs happen.

Author’s Side Note: If you want to learn something very interesting about venture capital, see how little press and discussion there has been about how wrong all of them were about self driving cars. They never admit their failures.

Not to mention a huge problem that is finally getting noticed. See, it takes a massive amount of power to run these huge data centers filled with NVIDIA cards. One data center uses about the same energy as 50,000 homes. And what does all that energy produce? Carbon dioxide. Google’s CO2 footprint has gone up 48% in the last 2 years thanks to its investments in AI. Many of the same companies that are massively increasing their emissions are the same companies that call the most for net zero protocols.

That is to say, almost all of the AI industry has two huge problems. First, their enormous use of energy clashes directly with their stated goals of reducing emissions. So far, very few people have called them out on massively increasing pollution while we all drink from paper straws. Second, the demand for energy is about to skyrocket and no one really knows where it’s going to come from. While of course they’d prefer renewables, the install base of AI data centers already on order are going to require 10 times the energy currently produced by solar and wind in the US…in the next 5 years.

Yeah, that isn’t going to happen.

So, what is happening? A massive resurgence in nuclear. Shocking how a power source so enormously hated by so many is suddenly in vogue when profit calls for it. The good news is nuclear is actually incredibly safe but almost no new plants have been created since regulations make them so expensive (while China is building nuclear plants as fast as they can). So of course Congress recently passed a bill getting rid of a bunch of red tape to make it easier for not only large energy companies but a huge number of small nuclear power startups designed just to power data centers bring potentially hundreds of new reactors online as quickly as possible.

And not unsurprisingly, nuclear stocks have been soaring while Bill Gates, Jeff Bezos, Sam Altman, Microsoft, and Google all are investing billions into new nuclear plants and companies.

I can’t say I’m really surprised the most obvious solution to zero emission power is suddenly popular once every company needs it to make money. Microsoft, Google, et. al. are not even pretending they can do this with anything other than nuclear. Although I find it extremely distasteful about how we are made to feel bad driving a gasoline powered car to visit family when companies are never shamed for massively increasing their emissions for AI. It goes to show how captured the media is by corporations that they are never called out.

So, time will tell if I’m right or not but barring a few incredible leaps in technology, LLMs will not usher in the future. I’m sure there are lots of people working on very creative new ideas but I see AGI arriving somewhere between 2030 and 2040, not sooner as many predict. I also worry that the price of electricity is going to skyrocket since keeping up with the baseload power necessary for all this new compute investment would require a miracle.

And for the record, it’s not AI. It’s not intelligent. It’s a very sophisticated chat bot. Until that changes, it’s not nearly as big a deal as everyone makes it out to be.