I have had a curious thirty days.
You know that old wedding saying, “something old, something new, something borrowed, something that scales three orders of magnitude?” I mean, that is basically how it goes, right? What I am seeing out there is kind of like that.
I have had two meetings within the space of one month of each other that have demonstrated technology advances for LLMs that will unlock multiple orders of magnitude of context window scalability. One of them was with a complete stranger, and one of them was with a friend of almost thirty years.
In the case of the friend, this was one of the people who showed me the BREW SDK from Qualcomm and sent me down the path of being one of the first ten mobile developers in North America to ship games to carriers.
I bring that up because I made a post on LinkedIn about the way that mobile adoption happened. It started slowly, and then one day it was everywhere. This was a journey of nearly a decade, and it was constrained by Moore’s Law, among other things.
But what happens when you don’t need to conform to something like Moore’s Law? It was one of the reasons that mobile phone adoption followed a slowly rising curve over eight years.
Instead, we have concurrent waves of innovation happening where the gains will not be slowly rolled out in n log n improvements per unit time, but one day, boom, your LLM vendor drops trou and there is a thousand times as much junk as there was before.
What this led me to do is come up with a corollary to Moore’s Law. The only right thing to do is call it Lesse’s Law:
“Sometimes technology advances will leap forward unbounded, resulting in hyper-incremental shifts in habits and livelihoods, in ways we cannot predict.”
I was recently in a roomful of people trying to explain the pace of advance we are about to see. It is like we went from pushing a slab of dirt covered with dried beans to feed the people in the cave to hauling tonnes of beans across the country on a maglev train to feed an entire city. Almost no other metaphor comes close to describing the leap forward we are about to take.
While these tools do not solve 100% of all the problems we experience in our day-to-day lives or jobs, they will come up with really good improvements and optimizations for much of our work, up to eighty percent, according to one of the studies on what kind of work LLM technology is good at.
I wish I still had the link to that study today.
This is not some kind of cheering. This is not some kind of Zoomer dooming either.
This is just a warning that it might be worth wearing your seat belt for the next couple of years.
Things are turning out wildly different from how I thought they would just three short years ago. And that will continue to change over the next fifteen hundred days.
Thank you for fastening your safety belt.
