Comparative Memory Systems in Brains and LLMs: A Functional Analysis
A recent paper on arXiv (2608.12377) explores the functional similarities between memory systems in biological brains and large language models (LLMs). It posits that, despite their distinct anatomical structures, both can be analyzed through common inquiries: the representation of memory-related information, the recovery of broader associations from partial cues, the updating of new information, and the perturbation of memory-related states. In biological contexts, these inquiries involve synapses, neuronal groups, hippocampal-cortical interactions, and plasticity, while in LLMs, they relate to weights, activations, context windows, retrieval mechanisms, and external storage. The study emphasizes functional and experimental aspects rather than anatomical ones. For further details, the paper can be accessed at https://arxiv.org/abs/2608.12377.
Key facts
- Paper ID: arXiv:2608.12377
- Announce Type: cross
- Compares memory systems in brains and large language models (LLMs)
- Shared functional questions include representation, cue recovery, writing/updating, and perturbation
- Biological systems span synapses, neuronal ensembles, hippocampal-cortical interactions, and plasticity
- LLMs span weights, activations, context windows, retrieval systems, and external stores
- Human studies show sparse concept responses, temporal binding, rapid association formation, episode-specific coding, and recall-related reactivation
- Rodent studies offer selective causal access to learning-related ensembles
Entities
—