A bank has fine-tuned a large language model (LLM) to expedite the loan approval process. During an external audit of the model, the company discovered that the model was approving loans at a faster…

Question

Options

• AInclude more diverse training data. Fine-tune the model again by using the new data.
• BUse Retrieval Augmented Generation (RAG) with the fine-tuned model.
• CUse AWS Trusted Advisor checks to eliminate bias.
• DPre-train a new LLM with more diverse training data.

• A
  72% (13)
• B
  6% (1)
• C
  17% (3)
• D
  6% (1)

Explanation

Explanation

Re-fine-tuning the existing model with more diverse, balanced training data directly addresses the root cause of the bias - skewed representation in the training set - while reusing the already fine-tuned model as a starting point, making it far more cost-effective than starting from scratch. Option D is wrong because pre-training an entirely new LLM requires enormous computational resources and cost, which is unnecessary when the base model only needs a targeted correction through fine-tuning. Option B is wrong because RAG is designed to augment a model's knowledge with external documents at inference time, not to correct systematic bias baked into model weights during training. Option C is wrong because AWS Trusted Advisor is an infrastructure and cost-optimization tool focused on cloud resource best practices - it has no capability to detect or eliminate AI model bias.

Memory Tip: Think of it as "fix what's broken at the source, cheaply." Bias entered during fine-tuning → fix it during fine-tuning with better data, not by rebuilding everything (D) or adding retrieval Band-Aids (B) or using the wrong tool entirely (C). The key phrase is "most cost-effectively" - fine-tuning again is always cheaper than pre-training from scratch.

No community discussion yet for this question.