Opinion | Are LLMs setting up our political ideologies?
Summary
LLMs are increasingly shaping how people seek and absorb information, raising concerns that their fine-tuning and human feedback may embed political leanings. A study cited in the article finds many conversational models tilt left of center, while base models appear closer to neutral.
Key Points
- Large language models have moved from simple chatbots to widely used information tools, often replacing search engines as the first stop for curiosity.
- A study published in PLOS ONE tested 24 conversational LLMs and found that most leaned left of center across multiple political orientation instruments.
- The article argues that political bias is often introduced during fine-tuning and human feedback, not necessarily built into the model architecture.
- For Nepal, the concern is that students and policymakers may unknowingly absorb answers shaped by hidden ideological assumptions from AI systems.
