Free GenAI Leader practice questions

AI is the broad field of machines doing tasks that normally need human intelligence; machine learning is the subset that learns patterns from data; generative AI is the subset of ML that creates new content. Natural language processing is the area of AI focused on understanding and generating human language.

Generative AI is a specialisation within machine learning, which is itself one approach within AI, so the circles nest AI > ML > gen AI. Inverting the order fails because gen AI depends on ML training. Treating gen AI and ML as non-overlapping siblings misses that every generative model is a machine learning model. Making NLP the umbrella confuses one application area (language) with the whole field; NLP overlaps gen AI where models generate text, but image generation is gen AI without being NLP.

1. Start from the broadest term: AI covers any machine performing intelligent tasks.
2. Narrow to ML: AI systems that learn from data instead of hand-written rules.
3. Narrow again to gen AI: ML models whose output is new content.
4. Place NLP as the language-focused area of AI, not as the umbrella over the others.

Exam tip: AI > machine learning > generative AI; NLP is the language area of AI, not the umbrella.

Google's AI-first approach means AI is the organizing principle of the company's research and products rather than a product line added later.

Google has invested in AI research for many years (its researchers introduced the Transformer architecture behind today's large language models) and has refined AI inside products such as Search, Maps, Photos and Gmail. The exam guide links this heritage and a commitment to future innovation to cutting-edge gen AI solutions: customers get models that come out of a continuing research pipeline. The distractors misread "first" as workload priority, a customer-before-internal release order, or a race to be first to market.

1. Recall that "AI-first" describes Google's own strategy, not a rule applied to workloads or release order.
2. Connect that strategy to years of research and product investment.
3. Link the investment to the depth and pace of today's gen AI offerings.
4. Discard readings that treat "first" as priority, sequence or market timing.

Exam tip: AI-first = AI at the core of Google's research and products for years, which is why its gen AI offerings stay at the cutting edge.

Hallucination is output that sounds plausible and confident but is factually wrong. It happens because a foundation model predicts likely text rather than looking facts up.

The assistant asserted a precise but false date about an old, well-documented event. That rules out the knowledge cutoff (the event is not recent), bias (no systematic skew toward a group) and an edge case (the request is routine). What remains is a hallucination, which Google Cloud addresses mainly with grounding and RAG, plus human review where errors matter.

1. Note what went wrong: a specific claim that contradicts a well-documented fact.
2. Check whether the event is recent (cutoff), a pattern against a group (bias) or a rare input (edge case); none applies.
3. Name the remaining limitation: a hallucination.
4. Link it to the remedy: grounding the model in verifiable sources.

Exam tip: Wrong but plausible-sounding facts about things the model could have known are hallucinations; grounding is the main fix.

The first step in choosing a gen AI solution is matching the output the business needs to a solution type: text, image, code, or personalized user experiences.

The deciding detail is the output: written descriptions, produced at volume, in a consistent voice. That is text generation. Image generation would create the product visuals, code generation serves developers, and personalization would matter only if each shopper had to see different copy; here each item needs one description for everyone.

1. Identify the output the business needs: written product descriptions.
2. Map written natural-language output to text generation.
3. Check whether the copy must differ per shopper; it does not, so personalization is not the core need.
4. Rule out the types whose output is visuals or software.

Exam tip: Name the output first: words map to text generation, visuals to image generation, software to code generation.

A foundation model is a large model pretrained on vast, broad data so it can be adapted to many downstream tasks. Reusing one pretrained model cuts the time, data and cost of building a separate model per task.

The stem's clue is reuse: one model serving several different tasks. Only a model pretrained on broad data has that generality. A small task-specific model trained on one team's labels is the old pattern foundation models replace. A large model trained on one domain's data for one task is big but still narrow. A fine-tuned model is a foundation model that has already been narrowed to one task, so it is the result of adaptation, not the reusable base.

1. Spot the requirement: one model reused across several tasks.
2. Ask what gives a model that range: broad pretraining on large, diverse data.
3. Separate the base model (foundation model) from a version adapted to one task (fine-tuned model).

Exam tip: Foundation model = pretrained once on broad data, adapted to many tasks.