From Short-Term to Long-Term Memory: The Context Window Evolution

The foundational large language models (LLMs) emerged with a profound limitation: a remarkably short memory span. This inherent constraint, often termed the “context window,” defined the immediate textual environment an AI could process and recall for generating its next token. In essence, the context window served as the model’s short-term memory (STM), encompassing the input prompt and the tokens it had generated so far within a