Building big language models demands access to a lot of compute power, high-quality data, and tools for managing bias. Another big problem is how data is used ethically and can be explained clearly by models.