AIF-C01 Applications of Foundation Models Practice Question

Which TWO of the following are valid methods to reduce the risk of foundation models generating harmful or biased content?

⚠ Common exam trap

AWS often tests the misconception that simply using a smaller model or disabling logging can reduce bias, when in fact these actions either have no effect or worsen the problem, whereas content filters and prompt engineering are direct, effective mitigation strategies.

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

✓

Use a content filter

Option B (Use a content filter) is correct because content filters act as a post-processing guardrail that screens both prompts and model completions for harmful, violent, hateful, or otherwise policy-violating content, blocking or redacting it before it reaches users. Option C (Apply prompt engineering to guide output) is correct because carefully crafted system prompts, few-shot examples, and instructions can steer the foundation model toward safe, neutral, and on-topic responses, reducing the likelihood of biased or harmful generations. Option A (Use a smaller model) is not a valid mitigation because model size does not determine safety or bias; smaller models can still produce harmful or biased content. Option D (Fine-tune the model on a biased dataset) would actually increase the risk by reinforcing biased patterns in the model's outputs. Option E (Disable all logging) does not reduce harmful content generation and instead removes the audit trail needed to detect, monitor, and remediate such issues.

Answer analysis

Option-by-option breakdown

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

✗
Use a smaller model
Why it's wrong here
Parameter count does not govern safety; smaller models still reproduce harmful or biased patterns learned from their training data. It tempts because smaller models are genuinely chosen when latency, cost or on-device memory are the constraints, not content risk.
✓
Use a content filter
Why this is correct
Content filters intercept prompts and completions at runtime, blocking harmful or biased output before it reaches users. This directly satisfies the stem's requirement to reduce risk from foundation models, since filtering operates on the model's generated content itself rather than on training data or access controls.
✓
Apply prompt engineering to guide output
Why this is correct
Prompt engineering steers a foundation model's output through carefully crafted instructions, system messages and few-shot examples, directly constraining the harmful or biased responses the stem asks you to reduce. It works at inference time without retraining, making it a valid mitigation alongside content filters and model selection.
✗
Fine-tune the model on a biased dataset
Why it's wrong here
Fine-tuning on a biased dataset amplifies the very bias the question asks to reduce, since the model learns and reproduces those statistical associations. Fine-tuning is the right technique when adapting a model to a legitimate domain or task, provided the training data is curated and representative.
✗
Disable all logging
Why it's wrong here
Disabling logging removes the audit trail needed to detect, investigate and filter harmful outputs, and does nothing to change what the model generates. Logging is correctly configured, not disabled, when monitoring invocations for abuse or compliance under Amazon Bedrock model invocation logging.

Go deeper

AIF-C01 Applications of Foundation Models Practice Question

Why each option matters

Option-by-option breakdown

Related to this question