Researchers warn that Large Language Models (LLMs) may exhibit implicit biases, even if they don't use explicit language, by inferring demographic data.