Generative AI Leader Google Cloud's Generative AI Offerings Practice Questions

Explanation: In Vertex AI Agent Builder, the 'Escalation' intent is specifically designed to hand off a conversation to a human agent when the agent cannot resolve a query. Option A is incorrect because slot filling collects parameters but does not transfer to a human. Option B is incorrect because a confirmation prompt only verifies user intent, not escalate. Option D is incorrect because a fallback intent handles unrecognized inputs but does not inherently route to a human; it can be configured to do so, but the standard mechanism for human handoff is the escalation intent.

Explanation: The latency issue for Australian users is likely caused by the vector search index serving from a single regional bucket in us-central1. When users in Australia query the index, the nearest endpoint must read the index data from a bucket far away, leading to high latency during peak hours. Option B solves this by creating a local bucket in australia-southeast1 and storing a copy of the index there. The endpoint in australia-southeast1 can then serve directly from the local bucket, reducing read latency. Moving to a multi-regional bucket (Option A) does not help because the bucket is still in the US, and Australia still has to fetch from a distant region. Option C would also work but is more complex and costlier than simply using a local bucket with the same endpoint. Option D addresses throughput but not the geographic distance causing latency.

Explanation: Combining human-in-the-loop review with automated scanning directly addresses the need to catch security vulnerabilities in AI-generated code. Human reviewers can identify logic flaws and context-specific risks that automated tools miss, while automated scanners provide consistent, rapid detection of known vulnerability patterns (e.g., OWASP Top 10). This layered approach is a best practice for production-grade code generation with Gemini Pro, as it mitigates the inherent limitations of relying solely on AI safety filters or static analysis.

Explanation: Setting safety thresholds to low (e.g., BLOCK_LOW_AND_ABOVE) for all categories blocks even low-probability harmful content, ensuring strict safety in a real-time chat application. Option A is incorrect because a high threshold (e.g., BLOCK_ONLY_HIGH) only blocks content with high probability of being harmful, allowing potentially harmful content through. Option B is incorrect because grounding with a safe knowledge base limits the knowledge source but does not filter the model's generated responses. Option D is incorrect because fine-tuning with safe examples reduces but does not eliminate the need for active safety filtering during inference.

Explanation: Batch prediction reduces cost because it processes multiple requests asynchronously in a single batch, allowing Vertex AI to optimize resource utilization and charge lower per-token rates compared to online (real-time) prediction, which requires dedicated infrastructure for low-latency responses.