Generative AI and Large Language Models (LLMs) AIE Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A media company wants to use an LLM to generate news articles. They are concerned about the model spreading misinformation and about the ethical implications of automated journalism. They have a team of editors who can review articles, but they want to minimize the number of articles that need significant corrections. Which strategy is most effective?

   • AUse prompts that require the model to base articles only on provided facts and to flag any uncertain information, then have editors review the flagged sections.
   • BUse a larger LLM that is known to produce more accurate articles.
   • CUse a discriminative model to classify news articles as reliable or unreliable.
   • DDeploy the LLM without any constraints and rely on editors to correct all errors.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A financial institution wants to use an LLM to generate reports from financial data. They have a pre-trained LLM and a small dataset of correctly formatted reports. They need the model to use specific financial terminology and follow a strict report structure. What is the most effective approach?

   • AUse a larger pre-trained LLM without fine-tuning, as it will automatically know financial terminology.
   • BUse the pre-trained LLM with a generic prompt to generate reports.
   • CUse a rule-based system that fills in a template with the financial data.
   • DFine-tune the LLM on the report dataset and use a prompt that specifies the report structure.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A data science team is investigating why their transformer-based LLM sometimes produces inconsistent translations for the same sentence when the word order is slightly changed. They suspect the model is not adequately capturing the position of words in the input. Which component of the transformer architecture is most directly responsible for encoding word order?

   • ASelf-attention mechanism
   • BPositional encoding
   • CFeed-forward network
   • DLayer normalization
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   Which of the following is a common mitigation strategy for LLM bias?

   • ACurating and filtering training data to reduce biased content
   • BUsing a larger GPU for inference
   • CIncreasing the model's temperature to 1.0
   • DDisabling the model's attention mechanism
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A team is deciding between a transformer-based LLM and an RNN for a real-time translation application. The application requires low latency and the ability to handle variable-length sentences. Which consideration is most important?

   • ARNNs are always faster than transformers for any sequence length.
   • BTransformers cannot handle variable-length sentences, while RNNs can.
   • CTransformer models can process tokens in parallel, which can reduce latency during inference compared to RNNs that process sequentially.
   • DRNNs require more training data than transformers.
   Reveal Solution Discussion Select an answer first