DRIFTLENS: Measuring Memory-Induced Reasoning Drift in Personalized Language Models

2026-07-02

AI Summary

This work investigates how incorporating user memory into personalized language models can induce substantial reasoning path deviations—even when outputs appear plausible—thereby compromising reliability. To address this, the authors propose DRIFTLENS, a novel framework that, for the first time, quantifies such reasoning drift in settings lacking ground-truth answers. By mapping reasoning steps to value categories and measuring trajectory divergence with and without memory via KL divergence, DRIFTLENS effectively distinguishes pragmatic noise from genuine shifts in reasoning. Experiments across four prominent large language models and ten user attribute types reveal moderate to large degrees of drift普遍存在. While integrating GRPO and DPO fine-tuning strategies partially mitigates this effect, the efficacy remains contingent on model architecture and reward design.

Abstract

Personalization changes what a model says to a user; we show that it can also change the reasoning trajectory used to justify the response. Modern LLMs personalize interactions by storing user attributes, preferences, and prior context, then injecting this information into future prompts. We study whether such memory reshapes reasoning on open-ended questions where no single ground-truth answer exists. To quantify this effect, we introduce DRIFTLENS, a ground-truth-free framework that maps each expressed reasoning step to a value category and measures divergence between a question's no-memory trajectory and its trajectory under injected user-attribute memory. We first validate that DRIFTLENS distinguishes content-free pragmatic noise from substantive reasoning changes. Across four LLMs and 10 user-attribute categories, including age, occupation, and disability, user-attribute memory induces medium-to-large reasoning drift above each model's pragmatic-noise floor, even when final answers remain fluent, on-topic, and plausible. We then evaluate GRPO- and DPO-based post-training methods for reducing drift. Both reduce drift, but neither uniformly dominates; effects on downstream capability, helpfulness, and instruction following are model-and reward-dependent. These results suggest that memory-induced reasoning drift is a measurable and only partly mitigated failure mode of personalized language models.

Problem

• reasoning drift
• personalized language models
• memory-induced
• user attributes
• LLM personalization

Innovation

• reasoning drift
• personalized language models
• memory-induced bias
• DRIFTLENS
• post-training alignment