About

I'm Jiří Klimeš, a software engineer by trade, and in my own time an ML researcher by stubbornness. One idea holds most of my attention: that today's models are artificial intuition, not intelligence. They recognize; they don't reason. Recognition is fast and cheap, and it runs out exactly where the interesting problems begin, at the boundary of the known. Past that boundary there is only search, and making search fast is the actual problem.

I taught myself to code out of boredom on a vacation: one eight-hour Python video, then a long stretch of tutorial hell. I climbed out by building things I actually wanted. A notification when my parents' phones hit the home wifi so I knew to look busy. An auto-ordering script for school lunches. A timetable app I wrote in Flutter in four days, with no prior frontend or Dart.

Then a friend asked whether neural nets would be hard, and we found out the slow way, from scratch, mostly without tutorials. We couldn't learn XOR (we were doing random mutation). We discovered a layer is just a matmul, tried genetic algorithms on Snake, then finally backprop. Digit classification, diffusion in NumPy, anomaly detection, language, and when the ARC-AGI prize was announced, three months on that. That's where it broke. I realized nothing I was building, LLMs included, could ever actually work, and deleted all of it. Getting my friend to see why is what sent me back to first principles, and I've been there since. light-search is what came out of that.

The plan, stated plainly so I can be held to it: build an assistant that compounds how fast I learn and build. Use it to help build real intelligence, then aim it at aging, the highest-leverage search in science: epigenetic reprogramming and the hallmarks, pushed far enough to reach rejuvenation and buy us time. Use the time to make the mind substrate-independent. Then go see the universe.

Email is best, and I read everything.