Stale, Misattributed, or Late: Where Personal Memory Fails Before Generation

Key points

• Personal memory for language agents is usually judged by whether the final an- swer is correct.1
• Missed merges leave stale values active, whereas false merges silently remove current values; four LLM key assigners achieve higher key re- call than a rule extractor yet produce lower clean-retrieval rates, and open-domain merge recall on LongMemEval never exceeds 0.062.1
• Misattribution survives va- lidity filtering: an entity posterior reduces same-name exposure on controlled data but cannot distinguish identically named speakers in LoCoMo.1
• These results argue for evaluating agent memory before generation, separating stored-state validity, identity resolution, abstention, and serving latency.1

Sources (1)

• [1]Stale, Misattributed, or Late: Where Personal Memory Fails Before Generation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 03:37 PM

  Using Personal Fact Memory (PFM) as a reference layer, we find that temporal validity is primarily a property of memory construction in our setting.

  Personal memory for language agents is usually judged by whether the final an- swer is correct.

Extractive summary: sentences quoted from the sources.