Question 1Fundamentals of gen AI
A commercial real-estate firm wants an assistant that answers questions about individual leases. A lease runs to several hundred pages, clauses cross-reference each other throughout, and every answer must cite the clause it came from. Volume is forty queries a day, and a few seconds of latency is acceptable. Which model selection criterion should dominate?
- A.
The lowest cost per token
- B.
Support for supervised fine-tuning
- C.
The fastest response time
- D.
A long context window
Show answer
Answer: D
Cross-referencing clauses across a several-hundred-page lease makes context window length the deciding model selection criterion.
- A. At forty queries a day token price is immaterial, so cost cannot outrank answer correctness.
- B. Fine-tuning shapes behaviour rather than storing changing facts, and leases would require constant retuning.
- C. The firm has already accepted a few seconds of latency, so response speed is not the binding constraint.
- D. Only a long context window lets the model reason across clauses that reference each other throughout a very long lease.