ResearchResearch paperEfficiency & Inference · Large Language Models1 source · Oct 6, 2026

When Old Facts Return: Re-Reads, Reverts, and the Limits of Temporal Memory

A memory system can retire an obsolete value and later restore it merely because the same old statement appears again.

Key points

  • A re-read of an old source and a genuine revert can produce the same observed sequence of values while requiring opposite current answers.
  • We study this ambiguity on 130 extractor-selected atomic transitions derived from software fixes.
  • In the ordinary transition condition, identity-based temporal memory reaches 98.5% model-judged accuracy with zero observed errors under a literal stale-value proxy.
  • Two supporting studies examine exposing retired history to the answer model and supplying current source for changed behavior.

Sources (1)

  • [1]When Old Facts Return: Re-Reads, Reverts, and the Limits of Temporal Memory
    arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 04:09 AM
    A memory system can retire an obsolete value and later restore it merely because the same old statement appears again.
    A re-read of an old source and a genuine revert can produce the same observed sequence of values while requiring opposite current answers.

Extractive summary: sentences quoted from the sources.

Related