Using this for semantic chunking feels like a great idea. Did you continue investigating that (after the second newsletter)?
Also, as I mentioned, now, trying out experiments in the video domain with some pretty cool results. Will post the details this week.