GCP Generative AI Leader Practice Question 3: Context Window Model Selection
Google Cloud Generative AI Leader · Test 1 · question 3 of 10
A legal operations lead is choosing a foundation model for contract review. Each master services agreement runs to several hundred pages. Reviewers want the model to flag conflicts between clauses that sit far apart, for example an indemnity cap on page 12 that is overridden by an exhibit on page 340. In a pilot, the model the team tried first could not accept a full agreement in one request.
Which model characteristic should weigh most heavily in the lead's selection?
Show answer and explanation
Answer:
The context window is how much information a model can hold and reason over in one request. Google's documentation compares it to short-term memory. Here the model has to connect page 12 with page 340 in the same pass, so the whole document must fit into the context at once. Context window size is one of the selection criteria Google lists for choosing a foundation model, along with modality, cost, performance, and security. Gemini models are built for long context and can take in very large inputs, such as long documents, in one prompt.
Option B sounds like customization, but it solves a different problem. Fine-tuning teaches a model a style, format, or task pattern from examples. It does not let the model read a document that is too large to fit in its context.
Option C breaks the task apart. If each clause goes into a separate request, no single request sees both the cap and the exhibit that overrides it. Cross-clause conflicts are exactly what would be missed. Cost still matters, but it cannot outweigh the one capability the task depends on.
Option D misunderstands temperature. Temperature controls how random the output is. It does not change how much text the model can accept, and more randomness is undesirable in legal review.
Reference: Long context