EngramaBench: A New Benchmark for Evaluating Long-Term Conversational Memory in LLM Assistants
EngramaBench is a new benchmark for evaluating long-term conversational memory across five query types including cross-space reasoning, temporal inference, and adversarial abstention. A controlled comparison of GPT-4o …
AI