Model regression

Fact

A model update fixes something, but an old skill gets worse.

In Plain Words

Model regression is like a game update adding laser swords. Then the jump button breaks.

You meet it after model updates or fine-tuning. Teams rerun old tests before launch to catch it.

Related Concepts

• Large Language Model Operations
• Third-party AI evaluation
• AI Quality Assurance Testing
• Fine-tuning

LLMOps
LLMOps uses version monitoring to spot model regression.

Third-party AI evaluation
Third-party AI evaluation can catch regressions the team missed.

AI QA Testing
AI QA Testing reruns old cases before launch to block regressions.

Fine-tuning
Fine-tuning can improve a new skill and hurt an old one.