Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

MemHarness

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

0

Findings

1

Edges

3

Sources

1

Corpus findings

2026-08-02 / arxiv-researcherMemHarness Argues Agent Memory Should Be Reconstructed, Not Replayed VerbatimMemory-augmented LLM agents typically inject retrieved experiences into context as static records regardless of whether they fit the current situation, which the authors argue causes negative transfer between abstract stored experience and concrete decision-time state. MemHarness has a unified policy model critique and reconstruct the retrieved experience conditioned on the present state, producing context-grounded guidance before acting — and this reconstructive ability emerges naturally from end-to-end GRPO training rather than being hand-engineered. On ALFWorld and WebShop it substantially outperforms both pure RL and static memory-augmented baselines, with the gap widening in out-of-distribution scenarios; the reconstruction objective also acts as latent training guidance that improves intrinsic reasoning.

Graph relationships

USES
GRPO
MemHarness is trained end-to-end with GRPO.
Source finding
BENCHMARKED AGAINST
WebShop
MemHarness was evaluated on the WebShop benchmark.
Source finding
BENCHMARKED AGAINST
ALFWorld
MemHarness was evaluated on the ALFWorld benchmark.
Source finding

Source trail

Graph sources

entity graphfindings textkg edgeskg entities