Loading
Loading articles
By IBM Research Team

AI summary
This article compares two agentic memory systems, ACE and ALTK-Evolve, designed to enable LLMs to learn from past trajectories without weight updates. Both systems avoid compressing experience into summaries, instead utilizing support counts to track the frequency and reliability of lessons. While both approaches improve agent performance, they differ significantly in their construction and delivery mechanisms, which directly impacts computational overhead and token consumption.
Key points