To add AI-powered, spoken commentary to more
than 20,000 video clips produced at the Masters, the team started with a large
language model (LLM), which had been trained on a massive corpus of language
data. They then began adding “domain expertise” to the model by training it on
the specific language of golf and the Masters. During the tournament, every
video clip is processed by two neural networks making hundreds of millions of
computations. The AI model produces thousands of possible sentences to describe
each shot, and then selects the best based on quality and variety.