Seed
A specific number passed to an AI model to make its generated output more deterministic and reproducible across multiple identical requests.
What it is
By default, AI generation has a degree of randomness. By providing a specific seed integer in an API request, the model attempts to make the output deterministic. If you send the exact same prompt with the exact same seed, the model will try to return the exact same completion.
When you would use it
You use a seed when you are testing your AI orchestration system and need the LLM to behave predictably rather than generating varied responses.
Common operations
- Writing automated unit tests for AI workflows that require consistent outputs.
- Reproducing a specific bug or edge case generation during debugging.
Related terms
Where this is taught
No learning path uses this term yet. Browse the Learning Atlas for guided sequences through related ideas.