What is Data contamination?
Data contamination is a data concept in AI development. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system.
Definition
Data contamination is a data concept in AI development. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system. The term is used in data discussions, where precision matters: two teams can use the same word to mean slightly different things. In this glossary, data contamination is treated as a practical concept rather than a formal specification.
Why it matters
Data terms describe how training, evaluation and retrieval corpora are collected, cleaned and governed. They matter because data quality and licensing often determine whether a model can be used in a regulated setting.
Example
Suppose you are comparing Phi-3.5 with another model for a production task. Data contamination is one of the factors behind the numbers you see: Phi-3.5 lists a 16,384-token context window, costs $2.81 per million input tokens and scores 76.2 on HumanEval. Understanding the term helps you decide which of those differences actually matter for your workload.
How to evaluate it
- Check whether data contamination is defined the same way in the documentation you are reading.
- Ask which measurement or trade-off the term is describing before comparing numbers.
- Separate vendor claims from independently observed behaviour on your own data.
- Write down the definition your team uses so reviews and evaluations stay consistent.
Related terms
Related guides
FAQ
What does Data contamination mean?
Data contamination is a data concept in AI development. It is part of the vocabulary teams use when they compare models, design prompts or review an AI system.
Why does Data contamination matter?
Data terms describe how training, evaluation and retrieval corpora are collected, cleaned and governed. They matter because data quality and licensing often determine whether a model can be used in a regulated setting.
How is Data contamination used in practice?
Suppose you are comparing Phi-3.5 with another model for a production task. Data contamination is one of the factors behind the numbers you see: Phi-3.5 lists a 16,384-token context window, costs $2.81 per million input tokens and scores 76.2 on HumanEval. Understanding the term helps you decide which of those differences actually matter for your workload.