You changed a prompt. You tried it once. It looked better. You shipped it.
That isn't testing, it's guessing. One good answer proves nothing — you've watched the same model nail a question and fumble it a minute later.
Write 20 real examples. Run them on every change. Count how many pass.
Vibes tell you it feels fine. Evals tell you it broke.
Hope this helps someone.
About me : I am a staff product analyst having interest in Product, Analytics, ML, GenAI, DE, Physics, Art, Literature. You can take mini AI course or read my mini blogs here : https://aisimplified.live
Top comments (0)