AIF-C01 Applications of Foundation Models Practice Question

A media company uses Amazon Bedrock to generate personalized news summaries for its subscribers. The model occasionally produces summaries that include outdated facts from its training data. The company wants the summaries to reflect only the most recent articles from its internal content management system (CMS). The CMS exposes a REST API that returns the latest articles. Which approach should the company take to ensure the generated summaries are grounded in the latest articles?

⚠ Common exam trap

The trap here is assuming that fine-tuning or retraining the model is the only way to update its knowledge, when in fact retrieval-augmented generation via agents or knowledge bases is the appropriate pattern for dynamic, up-to-date grounding.

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 Amazon Bedrock agents with an action group that calls the CMS REST API to fetch the latest articles, and include those articles in the prompt.

The requirement is to ground generated summaries in the most recent articles from a CMS. Amazon Bedrock agents with action groups can call the CMS REST API at inference time, fetch the latest articles, and pass them as context to the foundation model. This retrieval-augmented approach ensures factual recency without retraining. Other options either alter generation parameters or involve static training methods that cannot dynamically incorporate fresh content.

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 foundation model using Amazon Bedrock continued pre-training with a small set of recent articles.
Why it's wrong here
Continued pre-training is a form of fine-tuning that adjusts model weights, but it is not designed for frequent, real-time updates and does not query the CMS at inference time. It would not guarantee that every summary reflects the absolute latest articles. This approach is too slow and static for a dynamic news summarization use case, and it risks overfitting to a small sample.
✗
Retrain the foundation model on the CMS articles daily using Amazon Bedrock custom model training.
Why it's wrong here
Amazon Bedrock custom model training is not a daily, low-latency operation and does not automatically ingest data from a CMS REST API. Retraining a foundation model is expensive, time-consuming, and unnecessary for grounding outputs in fresh content. The scenario requires dynamic retrieval of recent articles, not model retraining. This option misinterprets the purpose of custom model training as a real-time data integration mechanism.
✓
Use Amazon Bedrock agents with an action group that calls the CMS REST API to fetch the latest articles, and include those articles in the prompt.
Why this is correct
Amazon Bedrock agents can be configured with action groups that invoke external APIs, such as the CMS REST API, to retrieve up-to-date content. The retrieved articles can then be inserted into the prompt as context, grounding the model's output in the latest facts. This is the intended pattern for dynamic, real-time grounding without retraining or fine-tuning the model.
✗
Increase the model's temperature setting to encourage more creative and current outputs.
Why it's wrong here
Temperature controls randomness and creativity, not factual recency. Raising temperature would make outputs more varied but would not inject new facts from the CMS. The model would still rely on its training data, which may be outdated. This option confuses a generation parameter with a data-retrieval mechanism, failing to address the grounding requirement.

Go deeper

AIF-C01 Applications of Foundation Models Practice Question

Why each option matters

Option-by-option breakdown

Related to this question