frontier models
Memory Rewrites Shrunk by Nearly 80%
MGPO credited each rewrite only for the downstream value it added beyond what the old memory already supplied.
Summary
MGPO credited each rewrite only for the downstream value it added beyond what the old memory already supplied.
Memory Gain Policy Optimization treats persistent textual memory as a delayed credit-assignment problem. It measures a rewrite's marginal contribution to current and future utility instead of rewarding value inherited from information already stored. On document-level information extraction, the authors report improved extraction with average memory length nearly 80% lower than the initial policy before optimization. The learned policy also transferred across domains and downstream models without more training. These results are task-specific and do not show that every long-running agent should optimize memory the same way.
Why it matters
MGPO credited each rewrite only for the downstream value it added beyond what the old memory already supplied.
Limits and context
- These results are task-specific and do not show that every long-running agent should optimize memory the same way.
Key claims
MGPO credited each rewrite only for the downstream value it added beyond what the old memory already supplied.
Qualification: These results are task-specific and do not show that every long-running agent should optimize memory the same way.
Evidence: source-2026-09-30-016
Sources
- arXiv preprint 2609.37930arXiv · primary research
Corrections
No corrections have been recorded for this story.