Generative AI and Large Language Models (LLMs) AIE Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A financial firm wants to deploy an LLM to answer customer questions about investment products. They have a limited budget and a small compliance team. The LLM must not give incorrect financial advice. Which approach best balances cost, compliance, and accuracy?

   • AFine-tune the LLM on a large dataset of financial advice and deploy it without human review.
   • BDeploy the largest available LLM and rely on its superior reasoning to avoid mistakes.
   • CTrain a small LLM from scratch on the firm's internal documents to ensure accuracy.
   • DUse a pre-trained LLM with prompt engineering that restricts answers to a provided knowledge base and routes uncertain queries to human advisors.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A healthcare chatbot uses an LLM to answer patient questions. The team wants to reduce the risk of the chatbot providing harmful medical advice. They have a limited budget and cannot afford extensive human review. Which strategy is most effective?

   • AUse a smaller model because it will be less capable of generating detailed medical advice.
   • BFine-tune the model on a large dataset of medical Q&A pairs to improve its medical knowledge.
   • CUse prompt engineering to instruct the model to only answer from a curated medical knowledge base and to decline questions outside that scope.
   • DDeploy the model with a high temperature to make it more cautious and less likely to give definitive advice.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A research team wants to use an LLM to generate scientific abstracts. They have a pre-trained LLM and a small set of well-written abstracts. They need the generated abstracts to be factually accurate and to follow a specific structure. What is the most effective approach?

   • AUse a text summarization model that extracts key sentences from the source.
   • BUse a larger pre-trained LLM without fine-tuning, as it will automatically generate accurate abstracts.
   • CUse the pre-trained LLM with a generic prompt to generate abstracts.
   • DFine-tune the LLM on the abstracts and use a prompt that specifies the structure and instructs the model to only use provided facts.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A developer is explaining to a colleague why a transformer-based LLM can handle long documents better than earlier recurrent neural networks (RNNs). Which statement is accurate?

   • AThe transformer's self-attention mechanism allows each token to directly attend to all other tokens in the sequence, reducing the need for sequential processing.
   • BThe transformer's positional encoding is only needed for short sequences, so long documents are processed without it.
   • CThe transformer uses a larger vocabulary than RNNs, which is why it handles long documents better.
   • DThe transformer processes tokens one by one, so it can remember the entire document in a fixed-size memory.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A developer is debugging why a transformer-based LLM sometimes produces grammatically correct but semantically nonsensical sentences. Which component of the transformer is most directly responsible for understanding word relationships in context?

   • APositional encoding only
   • BThe softmax output layer
   • CThe tokenizer vocabulary size
   • DSelf-attention mechanism
   Reveal Solution Discussion Select an answer first