AI and LLMs · Senior · Question 53 of 75
What is the role of guard tests when upgrading the model behind an agent?
- 1Re-run trajectory and safety evals for new behaviours
- 2Skip testing, since newer models are always safer
- 3Test only that the API still returns status 200
Show the answer
Correct answer: Re-run trajectory and safety evals for new behaviours
Explanation
A more capable model may also take bolder actions. Safety and permission tests matter as much as quality.