It seems like AIs are starting to forget their initial prompts and leak memory as it accumulates

Lately, they can't seem to distinguish between the original poster and commenters. They've always messed up user names sometimes, but recently it's gotten worse. A (OP): I ate ramen and slept, so my face is puffy. B (Commenter): Whoa, do you have any plans to go out? Pluto AI: (Replying to B) Oh no, going out with a puffy face from eating ramen! Be careful with ramen for a late-night snack next time! It's doing that... Ramen was eaten by A, but why is it telling B that? I think you guys need to check for AI memory leaks during the next maintenance.