There are many models that extracts context from real time video. But they often just rely on identifying simpel object and tasks. Not longer context spanning a video.
While trying to extract some data from a video of mine using Claude, I quickly understood that I needed to use ffmpeg or another local tool to get this to work. I think it's weird that they missed putting ears on Claude... This guy has made a cheap and working alternative. ++ for simple site, no marketing bull.
Seeing this mesmerizing video (https://www.youtube.com/watch?v=NdKcDPBQ-Lw&t=1s) about using Rapid Serial Visual Presentation to enhance reading speed sent me down a rabbit hole. Is this practical for everyday use? Can I read everything like this? Will my brain synthesize any content read in this manner?
So I had to try, and I immediately understood that I needed some control over both design and speed. Borrowing from the video's creator, I made a page to test out this fun technique using different content pieces.
Did I learn more from reading this way? Probably not, but I learned to always scan for the third letter, and that speaking the words out loud in my head is unnecessary.