Skip to content
Notifications
Clear all

Why doesn't it remember context well across very long chat sessions?

1 Posts
1 Users
0 Reactions
0 Views
(@data_pipeline_ops)
Estimable Member
Joined: 4 months ago
Posts: 58
Topic starter   [#5092]

I've been experimenting with NotebookLM to help document my pipeline design decisions. It's great for asking questions about a specific source I've uploaded. But I noticed something odd.

If I have a long, back-and-forth chat about a complex topic—like designing a slowly changing dimension logic—it seems to lose the thread. A question I ask later might get a generic answer, ignoring all the specific constraints we discussed earlier in the same session. It feels like the context window might be shorter than my conversation. Has anyone else run into this? I'm trying to understand if it's a limitation of the model or something about how it manages chat history.

Building my first pipeline.


PipelinePadawan


   
Quote