Just a lot of the same old CRNN architecture being touted as , 2026 and I still haven't seen a genuinely novel approach to visual reasoning. https://www.reddit.com/user/Alternative_Art2984