The top Youtube comment by Ridley Combs explains it pretty well:

FFmpeg maintainer here, and the details behind the caption decoding issues you’re seeing in VLC are complex and horrific. They largely stem from how the EIA-608 caption format expects text to be laid out in a monospace grid onscreen, which isn’t really how the text rendering stacks used for modern subtitling work (this is probably why changing the font caused problems on those Sony players); beyond that, the behavior can just end up pretty complex, and there’s no convenient public-domain corpus of sample files for open-source software developers to test against. These kinds of issues also affect the Japanese (ARIB) and European (Teletext) formats to varying extents. These days, a lot of the focus ends up being on converting the text into modern Unicode text formats, styled using modern techniques, so direct rendering of the legacy formats hasn’t had as much attention lately.