Andrej Karpathy stated in an August 5 2017 post that gradient descent can write code better than humans as the screenshot resurfaces amid later AI coding tools

This was software 2.0 Hardware 2.0 is coming

@sattyyouneed bet

@sattyyouneed tbf he did work at the company that pioneered it lmao

@andrewarruda Bro that’s crazy

I mean is it ?

@sattyyouneed 2015:

@sattyyouneed Andrej karapathy😭😭🙏🙏

@sattyyouneed 9 years ago? No chance in hell I would understand what he meant back then

@sattyyouneed Self admittedly, he wasn't talking about the act of writing code, rather the act if growing an AI vs programming it symbolically.

This was a stake on the connectionst/ symbolic convo.

@sattyyouneed He’s always early! Nobody like him.

@sattyyouneed @tobi @andrewarruda was earlier

@shregupta89 Yeah visionary

@andrewarruda Followed u

@sattyyouneed insanely early - I wonder what he is actually working on now

@MythThrazz @sattyyouneed here :)

@sattyyouneed This is about Software 2.0, weights replacing lines of code, not LLM programming

@sattyyouneed 💪

@sattyyouneed 2017: “Gradient descent can write code better than you.”

2026: junior devs are shipping apps with AI copilots before learning half the syntax seniors spent years memorizing. 😭

Some people saw the wave early. Most of the industry laughed first.

@sattyyouneed 🤯

@beffjezos Energy-based transformers https://alexiglad.github.io/blog/2025/ebt/