Why Kalmantic
Roger Penrose defined my AI worldview for a long time.
Matrix came out during my undergraduate years. Fell into the rabbit hole. Strong vs weak AI. Gödel’s incompleteness theorem. Emperor’s New Mind. Can a machine actually think, or is it forever imitation?
That question led to my first computer vision startup. Hand-crafting SIFT feature vectors. Building classifiers pixel by pixel. Market wasn’t ready. Compute hadn’t caught up. Timing was off by more than a decade. The startup folded. Practically every computer vision company from that era folded. Nevenvision. Riya. Long graveyard.
Those lessons stay with you. Scale and compute win. The elegance did not matter.
Three years since ChatGPT and there have been inflection every six months. Things have shifted. Hard to fully explain, but it’s there.
To think differently, you have to go to the source.
Incubating Kalmantic applied AI research Labs.
The name comes from two mathematical foundations that shaped how I think. Kalman filter and Karhunen-Loève transform. One tracks dynamic systems through noise, finding signal when measurements are uncertain. The other finds the most efficient way to represent complex data, keeping what matters.
Find signal in noise. Keep what matters.
Where does value actually accumulate when AI reshapes everything every six months?
The world is obsessed with growth. Growth matters from the outside.
At some point the question changes. What is the terminal value of this business?
Watched this question destroy companies that couldn’t answer it. Startup grows 3x year over year. Raises a Series A. Raises a Series B. Then a foundation model release makes their core product a feature inside a larger platform. Growth was real. Terminal value was not.
Growth without having a chokepoint is just a treadmill. Find the chokepoint, you get a toll booth.
The entity controlling the chokepoint in any value chain captures the surplus. Technology shifts move where the chokepoint sits. Principle never changes. The address does. AI is moving the address every six months.
Spending a lot of time around the kind of ideas Karpathy has been pushing.
Publishing research and benchmarks. Open source code, research papers, books.
Just going back to the source.