Compaction beats full transcript in memory
Original title: Compaction, decay and reinforcement in a two-tier conversational memory: a paired evaluation against retrieval-augmented generation
Compaction scores 71.4 percent vs 64.6 percent with 34 percent fewer tokens on LoCoMo.
Two-tier conversational memory compared using LoCoMo dataset. Compaction stores atomic facts while retrieval-augmented generation keeps transcripts. Compaction wins on comparable work but loses on adversarial sets where baseline leads by 14.6 points.
- 512 questions where methods disagree
- Compaction at 71.4 percent vs 64.6 percent
- 34 percent fewer prompt tokens
- Lead falls to 2.0 points on adversarial set
- Judge bias ruled out by token overlap analysis
From the original article
1 min read at arxiv.orgA conversational agent that remembers has to decide what to keep. Retrieval-augmented generation keeps the transcript and searches it; the alternative is to compact each turn into atomic facts and search those.
By Osamudiamen Edogun · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.