Do LLMs dream of elephants (when told not to)? Latent concept association and associative memory in transformers

arXiv:2406.18400v1 Announce Type: new
Abstract: Large Language Models (LLMs) have the capacity to store and recall facts. Through experimentation with open-source models, we observe that this ability to retrieve facts can be easily manipulated by changing contexts, even without altering their factual meanings. These findings highlight that LLMs might behave like an associative memory model where certain tokens in the contexts serve as clues to retrieving facts. We mathematically explore this property by studying how transformers, the building blocks of …

abstract arxiv association capacity concept cs.cl cs.lg elephants experimentation facts language language models large language large language models llms memory observe open-source models recall stat.ml store through transformers type