Random Is Hard to Beat: Active Selection in online DPO with Modern LLMs

TL;DR

In the context of strong pre-trained language models, active preference learning offers minimal benefits over simple random sampling for data selection, with negligible improvements in proxy win-rates and no mitigation of capability degradation.

Contribution

This paper critically evaluates the effectiveness of active preference learning in large language models, showing it often underperforms simple random sampling in strong prior regimes.

Findings

Active preference learning yields negligible improvements over random sampling.

Win-rate improves even as overall model capability degrades.

Active preference learning does not significantly reduce variance or mitigate capability collapse.

Abstract

Modern LLMs inherit strong priors from web-scale pretraining, which can limit the headroom of post-training data-selection strategies. While Active Preference Learning (APL) seeks to optimize query efficiency in online Direct Preference Optimization (DPO), the inherent richness of on-policy candidate pools often renders simple Random sampling a surprisingly formidable baseline. We evaluate uncertainty-based APL against Random across harmlessness, helpfulness, and instruction-following settings, utilizing both reward models and LLM-as-a-judge proxies. We find that APL yields negligible improvements in proxy win-rates compared to Random. Crucially, we observe a dissociation where win-rate improves even as general capability -- measured by standard benchmarks -- degrades. APL fails to mitigate this capability collapse or reduce variance significantly better than random sampling. Our findings suggest that in the regime of strong pre-trained priors, the computational overhead of active selection is difficult to justify against the ``cheap diversity'' provided by simple random samples. Our code is available at https://github.com/BootsofLagrangian/random-vs-apl.