Zero-shot
When a model handles a task it was never specifically trained for, using only its general training. A zero-shot imaging model can flag a finding without task-specific examples for that exact use case.
A radiology group tries a general image-and-text model by typing a prompt asking it to flag pneumothorax on chest X-rays, without giving it a single labeled example. It might do well on large pneumothoraces and poorly on small apical ones, and its performance on the next finding someone tries can bear no relation to this one. Treat each finding as a separate validation with a few hundred labeled local cases, and confirm that the model's terms of use and regulatory status permit clinical use before any zero-shot output reaches a worklist.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →