hardMultiple Choice
Generative AI Leader Practice Question: A data scientist is fine-tuning a large language…
A data scientist is fine-tuning a large language model for a legal document summarization task. The dataset contains only 500 examples, and the model must not forget its general language capabilities. Which fine-tuning method is most suitable?
⚠ Common exam trap
Generative AI Leader often tests the misconception that 'fine-tuning' always means updating all weights; candidates miss that LoRA is a fine-tuning method, not a prompting technique, and confuse it with in-context learning.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Adapter-based fine-tuning using LoRA
With only 500 examples and a requirement to preserve general language capabilities, parameter-efficient fine-tuning via LoRA adapters is ideal. LoRA freezes the base model weights and trains small low-rank matrices injected into attention layers, so the model retains its pretrained knowledge while learning the legal summarization task. It also dramatically reduces GPU memory and training time versus full fine-tuning.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Retraining the model from scratch on the legal dataset
Why it's wrong here
Training from scratch discards the pretrained weights entirely, so the model cannot retain general language capabilities and 500 examples are far too few to learn language anew. It is tempting because it gives complete control over the resulting model, and would be correct with massive domain-specific corpora and no need for transfer.
- ✓
Adapter-based fine-tuning using LoRA
Why this is correct
LoRA freezes the pretrained weights and injects small trainable low-rank matrices into the attention layers, updating only a fraction of parameters. With just 500 examples, this drastically reduces overfitting risk and preserves the model's general language capabilities, unlike full fine-tuning which would catastrophically forget them.
- ✗
In-context learning with a few examples in the prompt
Why it's wrong here
In-context learning places examples in the prompt without updating any weights, so it performs no fine-tuning at all and cannot durably adapt the model to the legal summarisation task. It is tempting because few-shot prompting suits tiny datasets and preserves general capabilities, and would be correct for quick adaptation without training.
- ✗
Full fine-tuning of all model parameters
Why it's wrong here
Updating all parameters on only 500 examples causes catastrophic forgetting of the model's general language capabilities, directly violating the stated requirement. It is tempting because full fine-tuning maximises task-specific adaptation, and would be correct when large labelled datasets are available and preserving prior knowledge is not required.
Go deeper
Related to this question
About these practice questions
Courseiva writes every Generative AI Leader question from scratch — 1,008 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official Google Cloud exam blueprint
This Generative AI Leader practice question is part of Courseiva's free Google Cloud certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the Generative AI Leader exam.