OpenAI Engineers Don't Understand Their Code, But AI Does

Jordan Nanos from SemiAnalysis revealed that OpenAI engineers, who were reviewing their core code with them, did not understand line by line what it actually does. And that didn't matter because AI understood it.

"It's not garbage code because it works. But essentially, it's literally generated AI assembler that is just dumped into Gluon—a low-level programming language for cores that they built on top of Triton and that uses a very interesting programming model."

"The essence is that the guys who were reviewing this code with us clearly understood hardware very well, understood the concepts of the whole system very well, but at the same time had no idea what exactly this MLA core for DeepSeek, which they showed us, does."

"You can go through the code line by line, and there literally: no, I don't know, I don't know."

"But that doesn't matter. AI understands it, AI tests it, and the results are visible. It generates correct cores that work really well."

"And I think this is just another sign of what awaits us next."

"The code itself doesn't necessarily have to be something that a person must deeply understand and dissect if AI knows how to manage the movement of data and computational elements on the hardware you gave it."