Actually he recently founded an AGI company. But this post is about optimization more than AI, which he is definitely good at (though he doesn't claim to be the best).
For someone to optimize something they need to have a good domain understanding. I do not buy that someone who developed games before just becomes an expert in LLMs overnight.
Sorry, it's definitely not true that you need to be an expert in LLMs to optimize BLAS kernels. Also, he started working on AI at least five years ago, which is a long time in AI world, probably before Justine Tunney started actually. And he's been optimizing code his whole life.
I was part of the Google Brain team working on TensorFlow back in 2015. Please note that doesn't mean I understand how LLMs work. Although I do know how matrix multiplication works now. At least until I apply my focus to the next area of impact and forget everything I just learned.
