Ty for reading the paper! I completely agree! Assigning soft weights to the window based on context is a fascinating research area. This concept is similar to Ebbinghaus' forgetting curve, which emphasizes recency bias while requiring repeated exposure for long-term retention.