In the news
Compaction, decay and reinforcement in a two-tier conversational memory: a paired evaluation against retrieval-augmented generation
arXiv cs.AI · Published · 3 min read
In 30 seconds
- What happened
- Researchers compared compacting conversation history into atomic facts versus retrieving full transcripts for conversational AI, finding compaction wins on standard benchmarks but loses on adversarial questions.
- Why it matters
- Matters for engineers building conversational agents deciding how to store and retrieve conversation context to balance accuracy and token efficiency.
- Watch out
- Compaction excels when answers need specific values but fails when phrasing matters; performance gap narrows significantly on adversarial examples, suggesting limited real-world advantage.
- agent
- retrieval
- eval
The patterns behind this
- Memory Decay & Forgetting Policies
- Generative Agents Memory
- Agentic Context Engineering (Evolving Playbook)
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.