I developed a project where Twitch chat could control a Doom game instance by sending commands, creating a fun and interactive experience.

The optimization of rendering in Unreal Engine was a significant challenge, focusing on reducing the cost of rendering pixels to improve performance.

We didn't compromise on making things easy, and because I relied on a foundation of principles, it was scalable to the largest applications on the internet.

DeepSeek optimized its models for efficiency by using 8-bit floating point format for training, which significantly reduces memory usage without sacrificing performance.

I realized that if you have a source of time in a network, you can build a more efficient system that utilizes the entire bandwidth allocated to it, unlike traditional protocols that suffer from inefficiencies.

The video diffusion model compresses hundreds of millions of internet videos into a simulation, enabling robots to interact with objects in a dream-like space.

One of my early projects involved translating classical projective geometry concepts into geometric algebra, which simplified the mathematical complexity of computer vision tasks.

The techniques used to bypass classifiers often involve encoding mechanisms that obscure harmful outputs, making them less detectable.

I shot Desperado digitally in my house in 1994, which was unheard of at the time. I insisted on editing it myself because it was crucial for me to have that creative control.