What is zero-shot learning?

A zero-shot claim only means something with a statement of what the model had not seen. A great deal of what is described as zero-shot is closer to well-represented in training, and nobody can check without that statement.

Ask for a worked example on your own material, and the evidence needed to reproduce it.

We use “Zero-shot learning” only where a product mechanism or an evaluation method backs it up, and we say when the behaviour depends on how a deployment is configured.