aiqbench analysis
This research introduces CAP-TTA, a test-time adaptation framework that dynamically applies LoRA updates only when detecting high-bias prompts in narrative generation tasks. By leveraging an offline precomputed diagonal preconditioner, the method achieves faster convergence and lower latency compared to standard optimizers while mitigating catastrophic forgetting. The approach effectively reduces toxicity scores on out-of-distribution data without sacrificing model fluency or established debiasing performance.
Preconditioned Test-Time Adaptation for Out-of-Distribution Debiasing in Narrative Generation
arXiv:2603.13683v4 Announce Type: replace Abstract: Although debiased large language models (LLMs) excel at handling known or low-bias prompts, they often fail on unfamiliar and high-bias prompts. We demonstrate via out-of-distribution (OOD) detection that these high-bias prompts cause a distribution shift, degrading static model performance. To enable real-time correction, we propose CAP-TTA, a test-time adaptation framework. CAP-TTA triggers context-aware LoRA updates only when a bias-risk score exceeds a set threshold. By utilizing an offline precomputed diagonal preconditioner, it ensures fast and stable optimization. Across multiple benchmarks and human evaluations, CAP-TTA effectively reduces toxicity/bias score with significantly lower latency than standard optimization methods (e.g., AdamW or SGD). Furthermore, it prevents catastrophic forgetting, and substantially improves narrative fluency over state-of-the-art baselines without compromising debiasing performance.
Abstract:Although debiased large language models (LLMs) excel at handling known or low-bias prompts, they often fail on unfamiliar and high-bias prompts. We demonstrate via out-of-distribution (OOD) detection that these high-bias prompts cause a distribution shift, degrading static model performance. To enable real-time correction, we propose CAP-TTA, a test-time adaptation framework. CAP-TTA triggers context-aware LoRA updates only when a bias-risk score exceeds a set threshold. By utilizing an offline precomputed diagonal preconditioner, it ensures fast and stable optimization. Across multiple benchmarks and human evaluations, CAP-TTA effectively reduces toxicity/bias score with significantly lower latency than standard optimization methods (e.g., AdamW or SGD). Furthermore, it prevents catastrophic forgetting, and substantially improves narrative fluency over state-of-the-art baselines without compromising debiasing performance.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY)
arXiv:2603.13683 [cs.CL]
(or arXiv:2603.13683v4 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2603.13683 Focus to learn more arXiv-issued DOI via DataCite
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 · 217 words
Related coverage
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
AINTMA: Agentic AI Architecture for Autonomous Test Management with Generative Intelligence, Secure Cloud Communication and Adaptive Quality Analytics
arXiv · Jul 24, 2026
Always ask your agent to be honest
LessWrong · Sep 16, 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · Oct 2, 2026