Clean, well-labeled datasets used for machine learning are partitioned into three subsets: Training sets, Validation sets, and Test sets. As your team is doing this, what's the best way to split up this data?
Correct Answer:
B
CPMAI's glossary defines data splitting as "dividing a data set into subsets (e.g., training, validation, test) for model development and evaluation," typically achieved via random subsampling to ensure each subset is representative of the underlying distribution and to prevent sampling bias.
- [Trustworthy AI]
Your organization has just rolled out a new image recognition system and is asking all employees to use it. It was trained using images from the ImageNet test set. After a few weeks, users are finding the results are not as expected and are asking for visibility into all the aspects of what went into building an AI system. What area of Trustworthy AI is being addressed here?
Correct Answer:
B
In CPMAI??s Trustworthy AI framework, Transparent AI focuses on providing clear documentation of data sources, modeling approaches, evaluation methods, and deployment plans so that stakeholders can audit and understand how the system was built. The users?? request for ??visibility into all aspects?? of model development, training data, and test sets directly maps to the Required AI Transparency Considerations task early in the methodology .
You have been tasked at your organization to manage a large language model (LLM) project. Identify what LLMs are useful for. (Select all that apply.)
Correct Answer:
BCDEF
Large language models (LLMs) excel at generating, understanding, and manipulating text. According to the CPMAI Glossary:
Content summarization is a core NLP function: "the process of using AI/ML techniques to generate a concise overview of a larger body of text."
Machine translation: "the use of AI to automatically translate text or speech from one language to another."
Classification: LLMs can assign content to categories via fine-tuned classification heads (??classifier?? term), making them suitable for content categorization.
Code generation: As generative AI, LLMs can produce new content, including code snippets, by pattern learning from programming corpora ("generative AI" term).
Search quality improvement: LLMs can rephrase queries, expand keywords, and rank results to enhance search relevance. Though not explicitly detailed in the glossary, this capability derives directly from their generative and understanding strengths.
LLMs are not designed for pure process automation (option A), which is handled by RPA or orchestrators rather than by text-centric models.
You want to create a model to figure out if a customer would be likely to repurchase a certain item. The project owner doesn't want you to create anything too complicated, and you have a limited data set to work with.
Correct Answer:
B
The CPMAI Glossary defines a naive Bayes classifier as "a family of simple probabilistic classifiers based on Bayes' theorem with the assumption of feature independence," making it ideal for small or limited datasets where model simplicity and interpretability are priorities.
- [AI Fundamentals]
Your team has built a new robot that roams the halls at your organization and helps with various things such as small deliveries. However, you notice that many employees are opting not to use the robot. When you ask them why they tell you that the robot looks ??creepy?? and they would rather not interact with it. What??s going on here?
Correct Answer:
B
This reaction is a classic example of the Uncanny Valley phenomenon, where a nearly human-like robot triggers discomfort or eeriness in users because it sits in the valley between clearly robotic and convincingly human appearances. Although not explicitly named in the CPMAI glossary, addressing this user experience concern falls under Continuous Improvement and Respect for People, ensuring cognitive solutions are designed for positive user acceptance.