We report here a comprehensive analysis about the political preferences
embedded in Large Language Models (LLMs). Namely, we administer 11 political
orientation tests, designed to identify the political preferences of the test
taker, to 24 state-of-the-art conversational LLMs, both close and open source.
The results indicate that when probed with questions/statements with political
connotations most conversational LLMs tend to generate responses that are
diagnosed by most political test instruments as manifesting preferences for
left-of-center viewpoints. We note that this is not the case for base (i.e.
foundation) models upon which LLMs optimized for conversation with humans are
built. However, base models' suboptimal performance at coherently answering
questions suggests caution when interpreting their classification by political
orientation tests. Though not conclusive, our results provide preliminary
evidence for the intriguing hypothesis that the embedding of political
preferences into LLMs might be happening mostly post-pretraining. Namely,
during the supervised fine-tuning (SFT) and/or Reinforcement Learning (RL)
stages of the conversational LLMs training pipeline. We provide further support
for this hypothesis by showing that LLMs are easily steerable into target
locations of the political spectrum via SFT requiring only modest compute and
custom data, illustrating the ability of SFT to imprint political preferences
onto LLMs. As LLMs have started to displace more traditional information
sources such as search engines or Wikipedia, the implications of political
biases embedded in LLMs has important societal ramifications.

大规模语言模型（LLMs）中嵌入的政治偏好的综合分析表明，当使用具有政治内涵的问题 / 陈述进行调查时，大多数对话型 LLMs 往往生成被大多数政治测试工具诊断为左倾观点的响应，这提供了关于政治偏好可能主要发生在 LLMs 的预训练后期、监督微调和 / 或强化学习（RL）训练阶段的有趣假设的初步证据。