Kind of wild how even the best models still struggle with keeping context straight over time. Definitely feels like a big challenge if we want these things to hold real conversations.