Everything AI Is Old Is New Again
Corporate lessons from yesteryear.
It is a time of change, of crisis. An extraordinary new technology is spreading; ordinary people dismiss or underestimate it, but those who work with it know it will transform our world. In my homeland of Canada, Prime Minister Trudeau is on the verge of departure, in the wake of scathing criticism of of his handling of deficits, productivity, and technology research, among many other issues. Meanwhile, the world is unnerved by saber-rattling from Moscow, and while Israel has recently withdrawn forces from within Lebanon to a border “security zone,” the Middle East remains extraordinarily tense and fraught.
The year is …
wait for it …
1983.
The political parallels are amusing … but the technological parallels are genuinely remarkable. Just as in the early days of computing, we have a behemoth that invented much of the basic technology, but has fallen behind in its sclerosis; an aggressive upstart seizing hearts and minds worldwide in a whole new way; and a smaller, more stylish upstart that seems especially beloved by tech hipsters and artisans.
Back then we had the operating system wars. Now we have the chat wars:
ChatGPT is Microsoft: the best-known, the people’s choice, the first in a critical new category, the de facto standard.
Claude is Apple: purists will tell you it’s definitively better, and in many cases they’re right, and it’s the overall choice of the educated classes, but nowhere near as successful when it comes to the masses.
Gemini is IBM: they invented half this stuff, but left it to others to actually implement and ship it, and are now scrambling to catch up. (Remember OS/2?)
Mistral’s Le Chat is Sinclair: sorry.
Llama is … and I know this sounds weird, it absolutely is … GNU/Linux. Right?! OK, Linux didn’t exist in 1983, but the GNU project began that year. And, I mean, the parallels are strong; open software that you can happily download and tinker with and tweak on your own computer! …It just feels odd that this time it’s being provided by a trillion-dollar company rather than a few MIT weirdos.
Note that I talked about the interfaces - ChatGPT, Claude, Gemini - rather than the models. The interfaces are what ordinary people use (or never use, in the case of Llama and Linux.) The models themselves map much more closely to the microprocessors of the 1980s. As such:
OpenAI is Intel: complex, sometimes needlessly and bafflingly so (Assistants API and Code Interpreter, anyone?) for historical reasons. Lots of options.
Anthropic is Motorola: they “led the wave of technologies that spurred the computing revolution in 1984, powering devices from companies such as Apple, Commodore, Atari, Sun, and Hewlett-Packard” … but they were no Intel.
Google is—again—IBM: just as Google’s Vertex AI will serve Claude, did you know that IBM teamed up with Apple and Motorola to build their own microprocessors, the PowerPCs? It did not go well, although they are active on the Mars rovers as we speak.
Llama is … uh. OK, these analogies, otherwise so compelling, kind of fall down here. There was no open source microprocessor, because microprocessors were/are insanely expensive to make. As are frontier models! … and yet Meta has, in its infinite generosity (hmmmmm….) bequeathed a semi-open one to us.
Looking at the above, you might conclude that the future is bright for OpenAI and Anthropic, dark for Google and Mistral, and unknowable for Meta. However. Google has something IBM didn’t; the example of IBM. (Which to be clear is still a very large and profitable company! …but does not dominate the tech industry as it did in 1983.)
It’s true Google published “Attention Is All You Need” and then apparently did nothing with it for a long time. But they also acquired DeepMind ten years ago, which could yet go down as the most prescient acquisition of all time. They’re trying to move as nimbly as they can. They have acquihired amazing talent for amazing prices. NotebookLLM was a bottoms-up hit. They still have a real chance of continuing to dominate! If only IBM had had their past selves as an example, too…
You’ll note I didn’t mention DeepSeek or any other Chinese models. That’s because, like Meta, there simply was no comparison back then; technology was an American business. That aspect of the future, too, seems impressively unknowable. Time isn’t entirely a flat circle. But it sure does rhyme sometimes.