[Submitted on 31 Aug 2026 (v1), last revised 26 Oct 2026 (this version, v3)]
Abstract:Large language models (LLMs) rely on static pretraining corpora, resulting in outdated knowledge over time. Existing approaches for evaluating knowledge edits either suffer from rapid contamination or rely on counterfactual edits that conflict with rigid existing knowledge. In this work, we propose a simulation-driven approach for evaluating and training LLMs on knowledge insertions. We introduce PARALLELEVENTS, a benchmark that simulates coherent trajectories of fictional yet realistic future events, enabling controlled evaluation without data contamination or conflicts with existing knowledge. We further propose SYNAPSE, a training framework that leverages model-generated data to integrate new knowledge into model parameters via instruction tuning. This synthetic pipeline enables scalable knowledge integration without costly human-curated data. SYNAPSE improves QA accuracy by 14.23 points over prior methods, indicating that simulation-based synthetic training yields robust, coherent knowledge insertion.
Submission history
From: Jonathan Zheng [view email]
[v1] Mon, 31 Aug 2026 18:06:35 UTC (6,829 KB)
[v2] Fri, 4 Sep 2026 17:29:17 UTC (6,829 KB)
[v3] Mon, 26 Oct 2026 23:01:49 UTC (5,896 KB)
