HPE AI Fundamentals (HPE0-V30) Certification Sample Questions

01. What does it mean to decompose a task in an agent design?
a) Breaking a goal into smaller sub-goals the agent can tackle in turn
b) Splitting the model across several accelerators to fit in memory
c) Dividing the document corpus into chunks before indexing
d) Separating the prompt into a system part and a user part

02. Which three projections of the input does an attention layer compute?
a) Token, position, and segment
b) Encoder, decoder, and classifier
c) Gradient, weight, and bias
d) Query, key, and value

03. A team is choosing tasks to pilot with a pretrained transformer, and wants ones the architecture suits without heavy customization.
Which two are natural fits?
(Select two.)
a) Guaranteeing arithmetic accuracy across large financial calculations
b) Summarizing long documents into short abstracts
c) Classifying free-text feedback into predefined categories
d) Providing a verifiable audit trail of why a specific decision was reached

04. What distinguishes extractive summarization from abstractive summarization?
a) Extractive selects existing sentences, while abstractive writes new text
b) Extractive requires a decoder while abstractive requires only an encoder
c) Extractive is only possible with fine-tuning, while abstractive works zero-shot
d) Extractive works on images while abstractive works on text

05. Which evaluation practice best reflects whether a RAG system is working?
a) Counting how many documents and chunks the index contains
b) Measuring only the fluency of the generated answers
c) Measuring each stage separately as well as end to end
d) Timing how long each query takes to complete

06. A team runs a small classical model over a modest table of numeric records. A batch completes in under a minute on ordinary servers. They ask whether moving it to GPUs would be worthwhile.
What is the soundest advice?
a) Move it, because GPU execution would also improve the model's accuracy
b) Move it, since accelerated hardware improves the throughput of any workload
c) Leave it, since classical models are unable to execute on accelerated hardware
d) Leave it, as the workload is too small to repay the added cost and complexity

07. Why is prompt injection a heightened risk in agent systems specifically?
a) Agents use larger prompts, which are harder for filters to inspect
b) Agents can act on injected instructions, not merely repeat them
c) Agents run without any logging, so injections go unnoticed
d) Agents cannot use a system prompt, so instructions cannot be prioritized

08. An insurer wants to extract policy numbers, dates and claimant names from free-text correspondence. Which task does this describe?
a) Named entity recognition
b) Machine translation
c) Sentiment analysis
d) Text summarization

09. What is the main objective a large language model is pretrained on?
a) Compressing documents so they occupy less storage than the originals
b) Retrieving the most relevant document for a supplied query
c) Predicting the next token given the tokens before it
d) Classifying text into a fixed set of labels supplied by annotators

10. Do orchestration frameworks include the language model itself?
a) Yes, but only for embedding rather than for generation
b) No, and they cannot be used with hosted model services
c) Yes, each ships with a model that the application uses by default
d) No, they connect to models the developer chooses