Sometime in the last year, someone created a satire complaint post on this subreddit that was sort of like a template. It was pretty funny and got a lot of engagement. I'm looking for it, and can't find it. Anyone remember it, have it in their upvote history?
Hi there.
Today, I had a Claude general chat session (newly formed this AM) do a very strange thing. It started answering questions in a fashion that was inarguably in response to a specific Project's instructions and/or context. Repeat -- the session was new, and in the general chat area. I have already reported this to Anthropic.
Settings->Capabilities:
Generate memory from chat history = off
Search and reference chats = on
I did not prompt it to search prior chats; even if it did, it would not explain the instructions compliance. From the very first question in the new session, it had this strange behavior.
Has anyone else begun noticing strange session context bleed today?
TY