What does the Foundry content filtering system check?
Why A is the answer
Both prompts and completions
Microsoft says the system runs both the prompt and the completion through classification models.
In detail
- The content filtering system checks both prompts and completions.
- Set filters to suit your audience.
Use case
An illustrative scenario, not a real incident.
A public app uses Foundry's content filtering. It checks what users send and what the model replies with.
See it in different ways
flowchart LR A["Prompt"] --> B["Content filter"] B --> C["Model"] C --> D["Content filter"] D --> E["Response"]
sequenceDiagram participant U as User participant M as Model participant K as Knowledge source U->>M: Question M->>K: Finds relevant content K-->>M: Passages M-->>U: Answer based on them U->>U: Still checks what matters
sequenceDiagram participant P0 as User participant P1 as Content filter participant P2 as Model P0->>P1: Prompt P1->>P2: If allowed P2-->>P1: Completion P1-->>P0: If allowed
Key point: Filters check prompts and completions.
Why the other options are wrong
- B. Only the images in a separate storage account
- It works on the model's traffic, not on a storage account.
- C. Only the prompt, before it reaches the model
- It also checks what comes back.
- D. Only the completion, after the model replies
- It also checks what goes in.