You know, I recently came across an interesting story about Andrey Karpaty. The guy is quite young by AI industry standards, but he has already left a significant mark on the technology. He holds six patents, most of which belong to Tesla. These are not just numbers on paper — they represent years of development that are now integrated into real systems.

Karpaty himself is an intriguing character. A Slovak who moved to Canada at 15, then went from academic to Google and OpenAI, and now is the lead autopilot developer at Tesla. He defended a dissertation on neural networks, taught at Stanford — all classic steps. But then something happened that changes people. He realized that theory must work in reality.

Karpaty’s patents cover serious topics: predicting 3D characteristics for autonomous driving, annotating videos with deep neural networks, creating data for machine learning. These are not abstract ideas but concrete solutions that Tesla has used for years. Five patents from the company, one from Google — and none for himself. His former employers will continue to benefit from his developments for a long time.

But the most interesting part is what Karpaty is writing and saying now. He introduced the term "vibe coding" to describe programming with AI assistants. He wrote an article, Software 2.0, where he develops the idea that neural networks will soon enable the creation of software that humans simply won’t be able to understand. It sounds ambitious, but looking at the pace of GPT and similar systems’ development, he’s not far from the truth.

What struck me most was his recent admission on social media. Karpaty said that for the first time in his life, he feels behind as a programmer. He compared AI tools to a powerful alien technology handed out to everyone without instructions. That’s from someone who wrote code for Tesla’s autopilot. If people like him are starting to feel overwhelmed, what about everyone else?

Right now, Karpaty is working on his startup Eureka Labs, creating educational videos about neural networks, and he’s also back at OpenAI. Interestingly, he tells schools: stop trying to catch AI in homework, because you still won’t understand it. The education system needs to change to match new realities. Karpaty clearly sees a future where AI is not just a tool but a partner in development.

According to Metaculus forecasts, just a couple of years ago, people thought true AI would appear in 2043. After GPT-4, that timeline shifted to 2028. Development is happening exponentially faster than expected. And people like Karpaty — they are at the center of this movement. His patents, ideas, observations — all building blocks of a new technological revolution. It seems we are really on the verge of something major.