Generative AI and Large Language Models (LLMs) AIE Practice Questions (Page 9)

Questions 41–44

1. 41application · medium

   A legal firm wants an LLM to summarize contracts using its own terminology and formatting standards. The firm has a large corpus of labeled contract summaries. Which approach best balances quality and cost?

   • AUse a rule-based text extraction system instead of an LLM.
   • BUse the pre-trained LLM with no modification and rely on its general summarization ability.
   • CTrain a new transformer model from scratch on the firm's contract corpus.
   • DFine-tune a pre-trained LLM on the firm's labeled contract-summary dataset.
   Reveal Solution Discussion Select an answer first
2. 42application · medium

   A company wants to automatically summarize long legal documents for their lawyers. They have a pre-trained LLM and a small set of example summaries. They need the summaries to be accurate and follow a specific format. What is the best way to achieve this?

   • AFine-tune the LLM on the example summaries and use a prompt that specifies the desired format.
   • BUse the pre-trained LLM with a generic summarization prompt.
   • CUse a discriminative model to classify the documents and then extract key sentences.
   • DUse a larger pre-trained LLM without fine-tuning, as it will automatically understand the format.
   Reveal Solution Discussion Select an answer first
4. 43foundation · easy

   Why is positional encoding used in transformer models?

   • ATo increase the model's vocabulary size
   • BTo reduce the computational complexity of self-attention
   • CTo convert text into a format suitable for GPU processing
   • DTo provide the model with information about the order of tokens in a sequence
   Reveal Solution Discussion Select an answer first
5. 44application · medium

   A news organization uses an LLM to draft article summaries. They notice the model sometimes produces summaries that contain facts not present in the original article. What is the most effective mitigation?

   • AUse a larger LLM with more parameters, as it will never hallucinate.
   • BIncrease the model's temperature setting to make the output more diverse and creative.
   • CUse a prompt that instructs the model to extract only information from the provided article and to avoid adding outside knowledge.
   • DRemove the article from the prompt and let the model summarize from its general knowledge.
   Reveal Solution Discussion Select an answer first