AIF-C01 Applications of Foundation Models Practice Question

Which TWO actions can help reduce bias in a foundation model’s outputs? (Choose two.)

⚠ Common exam trap

AWS AI Practitioner exam often tests the misconception that increasing randomness (temperature) or model size can inherently fix bias, when in fact these changes do not address the underlying data or prompt-level causes of biased outputs.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Fine-tune the model on a balanced, representative dataset

Option A is correct because fine-tuning on a balanced, representative dataset directly addresses bias by exposing the model to fair, diverse examples during training, which adjusts its learned weights and reduces skewed associations in its outputs. Option B is correct because careful prompt engineering with neutral wording avoids injecting biased framing or leading context into the input, so the model is less likely to amplify stereotypes or produce skewed responses. Option C is incorrect because restricting access to a subset of users is an access-control measure that does nothing to change the model's inherent biases. Option D is incorrect because increasing temperature only adds randomness to token sampling, which can make outputs more erratic rather than less biased. Option E is incorrect because a larger foundation model does not inherently reduce bias and may even reproduce or amplify biases present in its larger training corpus.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✓
Fine-tune the model on a balanced, representative dataset
Why this is correct
Fine-tuning on a balanced, representative dataset corrects skewed statistical associations in the training data, so the model's outputs reflect fairer distributions across demographic groups. This directly targets the root cause of bias rather than merely masking its symptoms at inference time.
✓
Use careful prompt engineering with neutral wording
Why this is correct
Neutral, carefully engineered prompts avoid loaded terms and leading framing that steer the model toward stereotyped or skewed outputs. This reduces bias expressed at inference time by constraining the context the foundation model conditions its response on, without retraining.
✗
Restrict model access to a subset of users
Why it's wrong here
Restricting access limits who can query the model; it does not alter training data or output distributions, so bias persists for permitted users. Access control suits least-privilege governance scenarios. Bias reduction requires intervening in data curation or output evaluation, not narrowing the audience.
✗
Increase temperature to add randomness
Why it's wrong here
Raising temperature only reshapes the probability distribution over tokens, so it varies wording rather than removing the skewed patterns learned from unrepresentative training data. It is tempting because temperature genuinely diversifies creative output, and it would suit brainstorming or generating varied marketing copy — not bias mitigation.
✗
Use a larger foundation model
Why it's wrong here
Scaling parameters does not remove bias encoded in training data; larger models can reproduce or amplify it. Larger models suit accuracy or capability demands, not fairness. Mitigation requires representative data, balanced sampling or bias evaluation, which model size alone does not provide.

Go deeper

AIF-C01 Applications of Foundation Models Practice Question

Why each option matters

Option-by-option breakdown

Related to this question