- Who it helps
- Researchers evaluating memory systems and developers handling changing user facts.
- Availability
- Public workshop preprint. This check established the abstract and date; a runnable reproduction was not verified.
The controlled comparison
The authors study a single-hop memory task. With a fixed prompt and four evidence statements, they replace an unrelated statement with an obsolete competing fact. Accuracy falls for both tested Qwen models, although the current fact remains available.
Obsolete facts can derail an answer even when the update is retrieved ↗
A useful evaluation question
Check whether a memory system answers from the active value when a previous value is also present. A retrieval metric that only checks whether the current statement appeared cannot establish that the final answer used it. The reported result concerns this benchmark and these configurations.
Obsolete facts can derail an answer even when the update is retrieved ↗
What you can do
- Read the paired evidence conditions before applying the result to a different memory system. Open reference ↗
Where the evidence stops
Abstract read; data, statistical analysis and experiments were not independently reproduced.
Sources read. Primary sources read November 1, 2026 UTC. No third-party installation, inference or complete workflow executed.
Sources & further reading
Paper · paper
Obsolete facts can derail an answer even when the update is retrieved ↗
Identifies the subject, documented capability or specific verification limit.
Research authors / arXiv · page checked 2026-11-01
Coverage note: Independent attention and end-to-end execution remain unverified.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
An October 26 study separates retrieving a current fact from using it correctly when an older version is also present.