I've been using Claude not for coding, but for analyzing writing and discussing stories. Ever since the upgrade to Claude 4, I've been facing a lot of frustrating issues. It keeps forgetting the first half of our conversation and even any documents I've sent. This is a real problem when I'm trying to analyze a story since it just loses track of everything I initially shared. I thought this might be due to the size of the documents I was uploading, but it happens regardless of the format. Has anyone else experienced this? Is there a way to fix it?
4 Answers
You’re not the only one! Issues with context have been around since Claude 3.5, just more noticeable with Claude 4. One trick you can try is using specific keywords like 'can you look back' to prompt it to reference earlier parts of the conversation. However, keep in mind it can sometimes misinterpret the context that way. Split your document into smaller chapters or sections to help with organization, even if it is a bit more work. It also seems like plain text formats function best, so keep that in mind too. Good luck with your analysis!
Have you considered trying NovelWriter? It might be an alternative for analyzing your stories, although I'm not sure how well it can handle analysis compared to Claude. Are you using it solely for story analysis, or would it help with generating some ideas?
Hey, I get where you're coming from! I had my own frustrations with it, but remember to keep experimenting with the format: sometimes just plain text doesn't mess with your flow as much as formatted documents. Hang in there, there are ways to make it work!
You're definitely not alone! What you're experiencing is actually due to the new Retrieval Augmented Generation (RAG) used in Claude 4. With this new system, it’s designed to remember only certain relevant fragments from the conversation rather than all details. So when it seems to forget parts of your earlier chats, it's not something you've done wrong. Here are a few tips to help you work around it: try breaking your analysis into smaller chunks, repeat important points throughout your conversation, or consider copying and pasting text directly instead of uploading it. Unfortunately, this is likely the new norm for Claude 4, and it seems Anthropic made some trade-offs for longer conversations at the cost of the model’s comprehensive understanding.
That's good to know, but sadly, none of those will help me. I can’t break things into smaller bits for my analysis. It's quite annoying when it starts mixing things up and making stuff up. Is this just a Claude 4 problem? Would older models work better for my needs?
Yeah, I've noticed similar issues even with older versions like 3.5. It wasn't as obvious back then, but the context window behavior has been part of the models for a while.
I mainly need something for analysis. I don't write new content with AI; I just want to interact with my existing stories. Claude was the best for that, but I'm open to experimenting.