Positive users say the study on LLM agent memory degradation matches their experience and nails the drift problem, while negative users criticize forcing summaries as teaching agents to gaslight themselves and prefer raw data instead.
4 comments with sentiment.