The world is being quietly rearranged by people who write very long documents.


The title they went with TSUBASA: Improving Long-Horizon Personalization via Evolving Memory and Self-Learning with Context Distillation Noisy translates that to

AI chatbots can now remember your preferences across months of conversations instead of forgetting after each chat


Researchers built a system called TSUBASA that lets AI chatbots hold onto and learn from your conversation history over long periods, instead of losing context after each session. This means an AI assistant could actually remember your preferences, past requests, and patterns—making it useful for things like project management, customer service, or personal assistants where continuity matters.
Every chatbot today either forgets you between sessions or gets slow and expensive when trying to remember everything. This paper shows a way to break that tradeoff: the AI learns from your history without requiring massive memory overhead. The catch is real—this is a research paper, not a deployed system. The benchmarks are synthetic, not user-facing applications. Whether this actually improves real personalization in the wild remains untested.
Watch whether companies building deployed AI assistants (customer service, productivity tools) start shipping memory systems that actually persist user preferences and improve recommendations over weeks or months, rather than resetting per conversation.

If you insist
Read the original →