Artificial Intelligence Governance Professional AIGP · Free Practice Question Medium
Question 4
A multinational corporation deploys a large language model (LLM) to automate customer service across multiple languages and regions. After several months, the support team observes that the model occasionally generates responses that are factually incorrect but highly plausible. What is the most likely explanation for this behavior?
- A The model's discriminative capacity is too high for the customer service use case
- B This is a sign of prompt injection attacks and not related to the model's architecture
- C LLMs only produce responses that are strictly based on their training data and never fabricate information
- D LLMs can generate fluent, plausible-sounding text that is not always grounded in factual data, a phenomenon known as hallucination
Reveal correct answer
Correct answer: D
Explanation
LLMs may "hallucinate" plausible but incorrect facts, underscoring the need for validation and human oversight in critical applications.A. Discriminative power isn't the root cause of factual errors in generative output.
B. Prompt injection is a separate security concern; plausible but incorrect outputs are a known LLM behavior.
C. LLMs can and do produce novel, unverified information.
D. LLMs predict likely sequences based on training data and sometimes create output that sounds correct but lacks factual accuracy.
Discussion
Think the marked answer is wrong, or have a better explanation? Share it below — comments appear after review.
You must be logged in to post a comment.
