aiqbench analysis
This study investigates the temporal validity of parametric memory in language models, demonstrating that current adapter-based methods fail to distinguish between stored facts and their expiration times without external retrieval or full history. The research highlights a fundamental separation between learning a fact and learning when to apply it, showing zero accuracy on time-sensitive queries for parameter-only approaches. These findings are critical for advancing reliable AI memory systems that can handle evolving user preferences and real-time event updates.
Engram Injection: A Study of Temporal Validity in Parametric Language-Model Memory
arXiv:2610.15882v1 Announce Type: new Abstract: A stored fact may stop being valid when a user changes a preference or an event replaces an earlier one. We study whether a language model can learn these changes in its parameters and answer correctly from only a question and the current time. Engram Injection stores facts in LoRA adapters and supplies the clock through FiLM conditioning on a frozen Qwen2.5-3B-Instruct model. We compare it with additive clock conditioning, prompting with the full dated history, and retrieving four dated records. We train six matched adapters across three seeds after fixing a sampler that could admit test-age examples into training. The training audit reports no such examples among 127,760 evaluated training inputs. On 62 pairs of questions asked before and after an update, at an elapsed age of seven days excluded from training, both adapter methods answer zero pairs correctly under the primary blinded assistant review with order code 1 held fixed. Full-history prompting answers 59/62 pairs correctly, and retrieval answers 62/62. Separate experiments show that adapters can store facts and that removable memory units can support local deletion. Combining fact selection with time-based version selection remains unsuccessful. These results separate learning a fact from learning when to use it.
Abstract:A stored fact may stop being valid when a user changes a preference or an event replaces an earlier one. We study whether a language model can learn these changes in its parameters and answer correctly from only a question and the current time. Engram Injection stores facts in LoRA adapters and supplies the clock through FiLM conditioning on a frozen Qwen2.5-3B-Instruct model. We compare it with additive clock conditioning, prompting with the full dated history, and retrieving four dated records. We train six matched adapters across three seeds after fixing a sampler that could admit test-age examples into training. The training audit reports no such examples among 127,760 evaluated training inputs. On 62 pairs of questions asked before and after an update, at an elapsed age of seven days excluded from training, both adapter methods answer zero pairs correctly under the primary blinded assistant review with order code 1 held fixed. Full-history prompting answers 59/62 pairs correctly, and retrieval answers 62/62. Separate experiments show that adapters can store facts and that removable memory units can support local deletion. Combining fact selection with time-based version selection remains unsuccessful. These results separate learning a fact from learning when to use it.
Artificial Intelligence (cs.AI)
arXiv:2610.15882 [cs.AI]
(or arXiv:2610.15882v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.15882 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 294 words
Related coverage
Asking For An Old Friend: Diagnosing and Mitigating Temporal Failure Modes in LLM-based Statutory Question Answering
arXiv · Sep 28, 2026
Benchtalks #3: We taught AI everything except how to learn
Snorkel AI Blog · Jun 25, 2026
Import AI 474: Platonic mindspace; TPUs in space; Zhipu starts an outer RSI loop
Import AI · Sep 28, 2026
Presentation: Engineering AI for Creativity and Curiosity on Mobile
InfoQ · Jul 21, 2026