AI is Becoming Normal
It’s Also Becoming Invisible
I’m not a fan of using AI. I’m not a fan of paying for it.
But my FitBit watch is now Google Health. It monitors my sleep and my exercise. It talks to me, and I talk back. Yes, I know it’s an AI, but I don’t care.
You will have noticed that a lot of the illustrations here now come from ChatGPT and Gemini. ChatGPT was ahead for a while, but Gemini is now better. (ChatGPT is also starting to demand money.)
I use both and take what works. Yes, I know they’re both AI, but a “cartoon” is more descriptive than a static image, and I don’t have to worry about Getty or some other copyright holder coming after me when I use something.
I don’t use CoPilot, or I pretend not to. But when it gives me those little blue squiggles underneath my text, I react. The AI inside the WordPress editor is also trying to teach me not to use so much passive voice. Sometimes it even succeeds.
There are hundreds of other examples I could cite from my daily life. Think about it and you can think of many, too, times when you don’t even think you’re using AI, but you are.
The point is that AI is entering all our lives slowly, on little cat feet, often without our realizing it. When we decide things are better with it, we accept it, even while we reflexively reject the technology and its implications.
In this way, again, AI is different from the PC, different from the Internet. The PC changed everything. The Internet made change fun. AI is more like global warming. It’s accretive. It sneaks up on us, the way a cat worms its way into our hearts.
The WDMing of AI
Om Malik has compared what’s happening to Wide Division Multiplexing (WDM), the use of colors to turn a single optical fiber into literally thousands of them. In my book on Moore’s Law, last updated in 2021, I told of how it killed Enron, which tried to make a market in bandwidth as it had once made one for natural gas. I compared it to selling kittens, and the assumption some foolish people have that they have economic value. (They don’t.) Since the fall of Enron, DWM has kept advancing, and can now run 51.2 terabits of data down a single fiber pair. The theoretical limit is even 12 times faster.
I think that’s what is going to happen with AI. This is really bad news for the data center boys. We have reached a point where AI companies must increase the price of tokens substantially to keep paying their bills. Companies that had been tokenmaxxing are acting like drug addicts. It’s especially bad for those using AI Agents.
I would argue that this is a good thing. It’s causing developers to focus on token efficiency, or token optimization. It’s creating new measurement terms like tokens per watt. It is even causing some people to question their own layoffs, because tokenmaxxed AI doesn’t save any money at all.
As with DWM, token prices can be made to fall. There’s a race going on, between token efficiency and token pricing. It reminds me a bit of how broadband companies tried to limit video downloads 20 years ago, capping streams to conserve their bandwidth. It’s also a phase.
Beyond the Crash
The rise of WDM, or DWDM (Dense Wide Division Multiplexing), was in the background of the Internet boom in the late 1990s, and the bust as well. It didn’t prevent what happened. The impact of it took time to manifest, as bandwidth prices fell and media learned how to make money from it. In that way the dot-bomb was healthy. It brought people into lower cost development modes, like open source, and gave the technology time to mature.
Cats multiply fast, but you still can’t rush gestation and birth.
Do what I do. Relax and say Om.
