引導式測驗 · 中等 等級
銀行關係經理人工智慧測驗
測試 RM 簡報工具如何使用權利過濾檢索、為什麼貸項憑證編號與敘述分開,以及下一步最佳操作的真正含義。
問題 1 的 8
In an RM assistant built on retrieval-augmented generation, when should access-control filtering happen?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- In an RM assistant built on retrieval-augmented generation, when should access-control filtering happen?
- After the model writes its answer, by redacting sensitive words
- Before retrieved text reaches the model
- Only when the banker exports the briefing
- Once a month during an audit
- Why does the guide recommend computing credit memo figures deterministically and inserting them as fixed values?
- Credit committees refuse to read AI-generated narrative
- It prevents the language model from inventing or miscalculating financial figures
- Deterministic numbers are required by suitability rules
- It lets the model skip reading financial statements
- What does a well-designed next-best-action system do?
- Automatically sells products to clients without the banker
- Replaces the banker's quarterly client reviews
- Presents options with reasons, leaving the banker to decide whether they fit
- Ranks bankers by how many products they sell
- Which constraint specifically prevents material non-public information from flowing between teams in an RM tool?
- Information barriers
- Contact frequency limits
- Model drift monitoring
- Data retention schedules
- What public example does the guide give of an AI assistant for advisors?
- A chatbot that approves mortgages for retail customers
- Morgan Stanley's GPT-4-based assistant that answers from internal research and procedures
- A central bank tool for supervising relationship managers
- A trading algorithm that executes client orders
- Why is a credit memo described as a control document?
- It controls the bank's marketing budget
- It sets the client's account permissions
- Credit committees rely on it to decide loans, so errors can lead to poor lending decisions
- It is automatically filed with regulators
- Besides propensity models, what do next-best-action systems typically combine to produce suggestions?
- Business rules such as eligibility, suitability and contact frequency limits
- Random selection of products to avoid bias
- Only the banker's sales targets
- Social media sentiment about the bank
- Why should next-best-action systems log each suggestion, its features and whether the banker acted?
- To calculate banker bonuses automatically
- To support model monitoring and conduct review
- To send the log to clients each month
- To train the model to ignore banker decisions