What is Zero-Shot Learning?
Zero-shot learning is the ability of LLMs to perform tasks using only instructions, without any examples. The model relies on knowledge transferred from pretraining to generalize to new tasks it has never explicitly seen.
Why It Matters
Zero-shot capability is what makes LLMs revolutionary—you can ask GPT-4 to classify legal clauses without showing examples, and it often works. This eliminates the need for labeled training data in many applications.
Zero-Shot Example
# No examples provided - just instructions
Classify the following text as POSITIVE, NEGATIVE, or NEUTRAL sentiment:
"The product works but the customer service was terrible."
→ NEGATIVE (model infers without examples)
When to Use Zero-Shot
- • Good for: Standard tasks (sentiment, summarization, translation)
- • Good for: When you don't have labeled examples
- • Good for: Exploratory/prototyping phase
- • Consider few-shot: Custom output formats
- • Consider few-shot: Domain-specific classifications
- • Consider few-shot: When consistency is critical