Trump's impact on reliability of LLMs — LessWrong