zero-shot
Without prior demonstration
Zero-shot describes a machine learning approach where a model performs a task without having been explicitly trained on examples of that task.
The model generalizes from related knowledge rather than learning from labeled data: The model achieved impressive zero-shot performance on languages it had never seen.
Compare with few-shot, where the model receives a handful of examples.
machine-learning
Conclusion reached through reasoning
- classificationSystematic grouping by traits
- predictionForecast of coming events
- generalizationDeriving principles from examples
- pretrainedPreviously taught skills
prompt · model · GPT · embedding · task · label · unseen · out-of-the-box
inference
Conclusion reached through reasoning
- classificationSystematic grouping by traits
- predictionForecast of coming events
- generalizationDeriving principles from examples
pretrained · prompt · model · GPT · embedding · task · label · unseen · out-of-the-box · emergent