AI memory framework MeMo skips LLM retraining

Researchers introduced MeMo, a modular memory framework that updates LLM knowledge without retraining the main model.

MeMo uses a small MEMORY model to learn new facts from generated Q&A pairs, while a frozen EXECUTIVE LLM calls it during inference.

The design works with both open- and closed-source models and aims to avoid RAG’s noise and fine-tuning’s cost and forgetting risks.

Tests showed MeMo could answer complex questions reliably, even when retrieval signals were noisy.