Meta’s Llama 3.1 405B, and OpenAI’s GPT-4o—were put to the test. Human participants had five-minute conversations with both a ...