Útmutatóhoz kapcsolódó kvíz · Közepes Szint
Why a Prompt Works in One Model but Not Another Quiz
Test prompt portability across models with matched cases, version records, and application-specific quality, format, and safety criteria.
kérdés 1 -ból 8
Miért viselkedhet ugyanaz a felszólítás két modellben eltérően?
Tekintse meg az összes kérdést8 kérdéseket
A válaszok és magyarázatok megjelennek a kvíz kitöltésekor.
- Miért viselkedhet ugyanaz a felszólítás két modellben eltérően?
- The words in the prompt change automatically
- All models use identical internal behavior
- Models can differ in instruction following, training, features, and API defaults
- A prompt is always model independent
- Az API-kompatibilitás garantálja a modell azonos viselkedését?
- Yes, compatible requests must produce the same text
- Yes, if the models share a context window
- No, because API compatibility is impossible
- No; request formats can match while outputs and supported features differ
- How can a team compare prompts across models fairly?
- Use a different test set for every model
- Use the same task examples, relevant settings, and scoring criteria
- Compare only marketing descriptions
- Run one prompt once
- Mit kell rögzíteni a hordozhatósági teszthez?
- Exact model identifiers, versions, settings, and test date
- Only the prompt’s filename
- Only average response length
- Nothing after deployment
- Which feature may differ even when APIs look similar?
- The fact that prompts contain text
- Whether a request has an input
- Structured-output support or tool-call behavior
- The concept of evaluation
- Hogyan kell értékelni a modell-specifikus prompt korrekciót?
- Assume it improves all tasks
- Change it in a controlled way and test against a holdout set
- Tune on every evaluation case
- Delete the baseline prompt
- Az egyszerű feladatokat portoló prompt automatikusan áthelyezi a biztonságra érzékeny feladatokat?
- Yes, if the wording is the same
- Yes, because safety rules are universal
- No, because simple tasks cannot be evaluated
- No; safety and tool behavior require their own tests
- What should a team do after a model or provider update?
- Rerun regression tests on the application’s requirements
- Assume previous scores still apply
- Only change the model name
- Remove the evaluation set