0
In a long chat, Claude can refer accurately to something I said near the beginning, then later miss another detail that was discussed just as clearly. It’s especially noticeable when I’m revising a document over many turns and expect it to keep track of earlier constraints.
Is this mainly a context-window limit, or can summarization and conversation structure affect what the model remembers? I’m curious what people have observed with long Claude conversations, particularly when the chat includes a lot of pasted text.
