Transfer Learning

Transfer Learning is a method of taking the ability learned in one domain and applying it to another. Think of it like someone who learned piano picking up an electric keyboard quickly — the underlying sense of the keys carries over.

The same happens in AI. Large models (foundation models) that have learned language ability by reading vast amounts of internet text can be retrained with just a little extra data for a specific field like medicine, law, or finance, and end up performing well with far less data and cost than training from scratch.

Most of what you now hear called "Fine-tuning" is really transfer learning in practice. When news reports say something like "a medical-specialized model was built on top of GPT," that's the capability of a general-purpose model being transferred. Thanks to transfer learning, companies no longer need to build their own AI from the ground up — they can simply add their own data on top of an existing model.