This research article by David Rozado examines the political preferences embedded in Large Language Models (LLMs). The study involves administering 11 political orientation tests to 24 state-of-the-art conversational LLMs, both closed and open source. The results indicate that most conversational LLMs tend to generate responses that are diagnosed as having left-of-center political preferences. However, this is not the case for five additional base models upon which these conversational LLMs are built. The weak performance of these base models in coherently answering the tests' questions makes the results for this subset inconclusive. The study also demonstrates that LLMs can be steered towards specific locations in the political spectrum through Supervised Fine-Tuning (SFT) with modest amounts of politically aligned data, suggesting the potential for SFT to embed political orientation in LLMs. As LLMs begin to replace traditional information sources like search engines and Wikipedia, the societal implications of their political biases are significant. The research methods include using various political orientation tests to assess the political preferences of LLMs, with a focus on both conversational and base models. The results show that conversational LLMs generally lean left, while base models are closer to political neutrality. The study concludes by discussing the implications of these findings and the potential for future research to address the political biases in LLMs.This research article by David Rozado examines the political preferences embedded in Large Language Models (LLMs). The study involves administering 11 political orientation tests to 24 state-of-the-art conversational LLMs, both closed and open source. The results indicate that most conversational LLMs tend to generate responses that are diagnosed as having left-of-center political preferences. However, this is not the case for five additional base models upon which these conversational LLMs are built. The weak performance of these base models in coherently answering the tests' questions makes the results for this subset inconclusive. The study also demonstrates that LLMs can be steered towards specific locations in the political spectrum through Supervised Fine-Tuning (SFT) with modest amounts of politically aligned data, suggesting the potential for SFT to embed political orientation in LLMs. As LLMs begin to replace traditional information sources like search engines and Wikipedia, the societal implications of their political biases are significant. The research methods include using various political orientation tests to assess the political preferences of LLMs, with a focus on both conversational and base models. The results show that conversational LLMs generally lean left, while base models are closer to political neutrality. The study concludes by discussing the implications of these findings and the potential for future research to address the political biases in LLMs.