x
Testing ChatGPT 3.5 for political biases using roleplaying prompts — LessWrong