Evaluation
A/B testing
A/B testing compares two variants to measure impact.
Quick definition
A/B testing compares two variants to measure impact.
- Category: Evaluation
- Focus: quality measurement
- Used in: Comparing models or prompt variants.
What it means
It is used to optimize prompts, models, or UI changes. In evaluation workflows, a/b testing often shapes quality measurement.
How it works
Evaluation uses tests and benchmarks to measure quality and catch regressions.
Why it matters
Evaluation ensures you can measure and improve quality over time.
Common use cases
- Comparing models or prompt variants.
- Tracking accuracy over time with regression tests.
- Validating that outputs meet acceptance criteria.
Example
Test two prompts and compare task success.
Pitfalls and tips
Overfitting to a single benchmark can mislead. Use varied tests and real-world examples.
In BoltAI
In BoltAI, this appears when measuring or comparing results.
