> but for what AI model?

Sentiment analysis. Working out what words lead to what outcomes, and then being able to predict on new data is super useful.

For coding or "AGI" no, its not useful. For building a text based (possibly image based) recategorisation system top class.