Human Memory Limits Make AI Better at Grammar (opens in new tab)

Researchers have developed "fleeting memory transformers" that integrate human-like memory decay and a 3-to-7 word echoic buffer into neural networks, proving that restricting an AI's context window forces it to prioritize abstract grammar over literal memorization, drastically improving low-data language learning.