I’ve been thinking about product development. We talk a lot about finding a problem and building a solution, but I’m not sure that’s how breakthrough products happen. Nobody thought they needed a car—they already had horses. It was only after cars existed that people realized what they had been missing. Great products don’t just solve problems; they unlock new ways of living.
Farsight Finds the Small Stuff—Ukraine’s Farsight Vision has upgraded its neural model so Ukrainian troops can detect the tiny signs that often matter more than the obvious ones:
wheel tracks, narrow trails, wires, poles, EW antennas, infrastructure elements and camouflaged objects.
The update adds more than 30 object classes to the FSV Platform, which turns drone imagery and video into orthomosaic maps and 3D models for planning assaults, FPV routes and vehicle movement, with access to the recognition system free for Ukraine’s Defense Forces.
Detection already works on orthophotos, and from May 15 it should also work directly on standard photos.
Farsight Vision: https://t.co/fzcvyzgTDc
https://t.co/z1rdG8fk8N
Get our daily insights: https://t.co/B67VPpa8Nk
@BRAVE1ua@Oboronka1 Amazing idea.
With the current pace of reinforcement learning — especially multi-agent RL — decentralized swarm behavior is no longer science fiction.
Most “defense AI” is trained on fantasy.
Flat maps.
Clean sensors.
Perfect conditions.
War is none of that.
Noisy data.
Broken geometry.
Occlusions.
Elevation.
Uncertainty.
If your model never learned on real 3D terrain,
it won’t survive real terrain.
Physical AI begins where slides end.
Technology itself is neutral.
But the architecture of decision-making is not.
Maybe the real question isn’t whether AI can deceive.
It’s this — Who controls the system that does the deceiving?
Yesterday I was talking with @vikyaremchuk about AI ethics.
If war is the art of deception — where are the limits for AI?
When autonomy scales faster than accountability, the real risk isn’t intelligence.
It’s loss of control.
AI in miltech shouldn’t remove responsibility from humans. The line isn’t “can we build it?”
It’s: who controls it, and who answers for it.
The future of miltech won’t be defined by bigger weapons.
It will be defined by smarter autonomy.
The winners won’t be those with the most data —
but those who can deploy robust Physical AI
in real 3D terrain, not flat maps.
Miltech is turning into the hardest applied AI field on Earth.
And that’s exactly why it’s fascinating.
@lexfridman I love your channel and I have great respect to you. Your father born in Kyiv, and you definitely should have relative there. So it’s feel strange that you don’t cover war until now.
Можно построить мельницы,чтобы не тратить драгоценные силы людей и животных на помол зерна. В хорошие времена усовершенствования могут казаться излишествами и без них можно прокормиться. Но в плохие они залог выживания.А когда снова наступают хорошие времена,инновации дают взлёт