Fine tuning a model for customer service for our specific app

Absolutely. That’s the power of these models, you can translate all text, images, video (24 images per sec), sound (spectrograms) into an embedding aka a vector aka a table of numbers.

With that now you can compare and relate all your data. GPT3 is a huge collection of embeddings and their relationships. Fine tuning it with your embeddings creates new relationships between these vectors, so now knows how to reply using your data.