Aspects of human memory and Large Language Models

TL;DR

This paper explores the memory capabilities of Large Language Models, revealing they exhibit human-like memory traits learned from training data, which reflect biological memory features influencing narrative structures.

Contribution

It demonstrates that LLMs develop human-like memory properties not inherent to their architecture but learned from textual data statistics.

Findings

LLMs show surprising similarities to human memory characteristics.

Memory traits in LLMs are learned from training data, not architecture.

Biological memory features influence how humans structure narratives.

Abstract

Large Language Models (LLMs) are huge artificial neural networks which primarily serve to generate text, but also provide a very sophisticated probabilistic model of language use. Since generating a semantically consistent text requires a form of effective memory, we investigate the memory properties of LLMs and find surprising similarities with key characteristics of human memory. We argue that the human-like memory properties of the Large Language Model do not follow automatically from the LLM architecture but are rather learned from the statistics of the training textual data. These results strongly suggest that the biological features of human memory leave an imprint on the way that we structure our textual narratives.