MIT and Meta Propose Self-Correcting Memory for AI Agents

1 reports archived1 independent sourcesupdated 11/13/2026, 08:39:57

Synthesis & Latest Updates

1 Sources Cross-Validated

A joint MIT and Meta paper introduces ReHarness, which revises an agent's previously retrieved notes using the outcome of each task instead of only appending new ones, preventing stale or over-broad lessons from misleading the agent. Across three benchmarks and three frontier models it averaged 0.240 normalized gain versus 0.168 for the best raw-trajectory baseline, while using roughly one-third the tokens of replaying full raw history on CL-Bench with Claude.

LATEST/A joint MIT and Meta paper introduces ReHarness, which revises an agent's previously retrieved notes using the outcome of each task instead of only appending new ones, preventing stale or over-broad lessons from misleading the agent. Across three benchmarks and three frontier models it averaged 0.240 normalized gain versus 0.168 for the best raw-trajectory baseline, while using roughly one-third the tokens of replaying full raw history on CL-Bench with Claude.

Hourly heat curve

5 fully observed hours

Now

14.2

Peak

15.410:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

Rohan Paul (@rohanpaul_ai)T2Nov 13, 08:39·71 pts
MIT and Meta Propose Self-Correcting Memory for AI Agents
- Outcomes are used to revise the notes that shaped them, rather than only appending new memory
- Averaged 0.240 normalized gain across 3 benchmarks and 3 models, beating the 0.168 raw-trajectory baseline
- Uses far fewer tokens than replaying full raw history, making it practically deployable
View Related Reports & Analysis Original Source