Data/ML/AITable-stakes
Few-Shot Learning
Also written as Zero-Shot Learning, One-Shot Learning
Getting a model to perform a task by showing it a few (or zero) examples in the prompt, instead of retraining it on a full dataset.
Think of it like
Like showing a new employee two or three worked examples of how you want a report formatted, instead of sending them to a full training course.
Junior or senior?
Junior sounds like
Adds examples to a prompt without a clear stopping point.
Senior sounds like
Can explain when they'd add examples versus fine-tune instead, and the cost/latency tradeoff of doing so.
Ask them
“How did you decide how many examples to include in a prompt, and when that stopped being enough?”