The Lex Fridman Podcast episode #496 features Jean-Baptiste Kempf and Karen Cunier discussing FFmpeg and VLC, the open-source technologies underpinning most internet video. They delve into the technical complexities of video compression, the volunteer-driven ethos of open-source development, and the critical role these projects play in democratizing multimedia and preserving digital heritage. The conversation also touches on recent controversies regarding corporate support for open-source and the future of ultra-low latency video for robotics.
Summarized by Podsumo
FFmpeg and VLC are the invisible backbone of internet video, powering platforms like YouTube, Netflix, and Chrome, handling nearly all video processing and playback, and downloaded billions of times.
Both projects are built by thousands of volunteers, prioritizing engineering craft and making complex multimedia accessible, with Jean-Baptiste Kempf famously refusing millions to keep VLC free and ad-free.
Extreme optimization through assembly language is crucial, with projects like David (AV1 decoder) featuring hundreds of thousands of lines of hand-written assembly, achieving 10x-60x speedups over C for real-time processing.
The discussion highlights significant challenges in open-source maintenance, including 'AI Slop' from security reports and disproportionate demands from trillion-dollar corporations on unpaid volunteers, leading to maintainer burnout.
The future of multimedia involves next-gen codecs (AV2, H.266), a shift towards royalty-free standards, and ultra-low latency platforms like Kyber, aiming for 4-millisecond glass-to-glass latency for remote control of robots and drones.
"FFMPEG is basically the low-level libraries for codec, to compressions and decompression, muxes and demuxes and filters. It's the core ethys and then you have a several tools which allow you to create a type of pipeline to process any type of video files."
"The world is a museum of passion projects. Everything out there is a passion project and open source multimedia and open source in general. You can just do that so much faster."
"Philosophically, what's important to realize is that we passed the time where hardware was going so much faster, right? We are at the end of the more low, we have limitation for AI, for memory. You need to go down in the stack and optimize more to get more power from what you have because our requests for power, CPU power, GPU power are exploding while the hardware is not exploding in speed."