Large language models show strong political bias: a simulated polling experiment | Not Hacker News!