Stayin' Aligned Over Time: Towards Longitudinal Human-LLM Alignment via Contextual Reflection and Privacy-Preserving Behavioral Data
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913091596320768 |
|---|---|
| author | Gebreegziabher, Simret Araya Sproul, Allison E Yang, Yinuo Chen, Chaoran Gómez-Zará, Diego Li, Toby Jia-Jun |
| author_facet | Gebreegziabher, Simret Araya Sproul, Allison E Yang, Yinuo Chen, Chaoran Gómez-Zará, Diego Li, Toby Jia-Jun |
| contents | Current human-AI alignment and evaluation methods for large language models (LLMs) often rely on preference signals collected immediately after an interaction. This practice implicitly treats preference as static, even though many LLM-mediated decisions unfold over time and may be re-evaluated differently after real-world consequences and observed outcomes. Therefore, we argue for a methodological shift from single-moment preference elicitation to longitudinal, context-situated alignment measurement. We present a methodological framework for collecting temporally grounded alignment signals by combining (1) in-situ preference capture, (2) context-triggered follow-up preference reflection, and (3) privacy-preserving behavioral traces that help interpret preference change. As an instantiation of this methodology, we introduce BITE, a browser-based system that detects consequential LLM interactions, prompts reflection across later decision points, and supports progressive, user-controlled consent for sharing behavioral data. Through a two week longitudinal deployment study with 8 participants, our approach surfaced differences between immediate and later user preferences in accuracy, relevance and other dimensions of the LLM output. Our findings highlight the limitations of single-moment preference datasets and underscore the importance of longitudinal methods for alignment evaluation in everyday use. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_04029 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Stayin' Aligned Over Time: Towards Longitudinal Human-LLM Alignment via Contextual Reflection and Privacy-Preserving Behavioral Data Gebreegziabher, Simret Araya Sproul, Allison E Yang, Yinuo Chen, Chaoran Gómez-Zará, Diego Li, Toby Jia-Jun Human-Computer Interaction Current human-AI alignment and evaluation methods for large language models (LLMs) often rely on preference signals collected immediately after an interaction. This practice implicitly treats preference as static, even though many LLM-mediated decisions unfold over time and may be re-evaluated differently after real-world consequences and observed outcomes. Therefore, we argue for a methodological shift from single-moment preference elicitation to longitudinal, context-situated alignment measurement. We present a methodological framework for collecting temporally grounded alignment signals by combining (1) in-situ preference capture, (2) context-triggered follow-up preference reflection, and (3) privacy-preserving behavioral traces that help interpret preference change. As an instantiation of this methodology, we introduce BITE, a browser-based system that detects consequential LLM interactions, prompts reflection across later decision points, and supports progressive, user-controlled consent for sharing behavioral data. Through a two week longitudinal deployment study with 8 participants, our approach surfaced differences between immediate and later user preferences in accuracy, relevance and other dimensions of the LLM output. Our findings highlight the limitations of single-moment preference datasets and underscore the importance of longitudinal methods for alignment evaluation in everyday use. |
| title | Stayin' Aligned Over Time: Towards Longitudinal Human-LLM Alignment via Contextual Reflection and Privacy-Preserving Behavioral Data |
| topic | Human-Computer Interaction |
| url | https://arxiv.org/abs/2605.04029 |