A chatbot gives confident answers about products launched after its knowledge cutoff. What addresses this?
Why D is the answer
Grounding it on current company data
Grounding gives the model current, verifiable sources to answer from.
In detail
- Grounding gives the model current company data.
- Grounding gives sources that can be checked.
Use case
An illustrative scenario, not a real incident.
A chatbot invents details about a new product. Grounding it on the current catalog gives it real facts to use.
See it in different ways
flowchart TD
A{"Where is the truth?"} -->|"Your company data"| B["Pre-built RAG with Agent Search or RAG APIs"]
A -->|"Public, current facts"| C["Grounding with Google Search"]stateDiagram-v2 [*] --> SourceOfFacts SourceOfFacts --> GoogleSearch: current public facts SourceOfFacts --> AgentSearch: company documents GoogleSearch --> GroundedAnswer AgentSearch --> GroundedAnswer GroundedAnswer --> HumanReview: high impact GroundedAnswer --> [*]
sequenceDiagram participant P0 as User participant P1 as Chatbot participant P2 as Company data P0->>P1: Question P1->>P2: Look up facts P2-->>P1: Current content P1-->>P0: Grounded answer
Key point: Grounding supplies current, verifiable sources.
Why the other options are wrong
- A. Raising the temperature for more creative answers
- Higher temperature can make it worse.
- B. Using a shorter prompt with fewer instructions
- A shorter prompt adds no facts.
- C. Removing the safety settings that limit answers
- Removing safety adds no facts.