Understand how different input types contribute evidence.
1. Modalities and tasks
Text, images, audio and video encode different information. Choose the modality that supports the user’s task rather than collecting everything a device can capture.
2. Joint interpretation
A model may combine instructions with visual or audio evidence. Conflicting inputs can produce uncertain results; design an explicit unknown or clarification path.
3. Perception limits
Generated scene descriptions are interpretations, not verified observations. Avoid inferring facts such as identity, intent or hidden conditions from ambiguous media.
Worked scenario
A photo shows a damaged box but not whether its contents are broken. The assistant records visible damage and leaves contents unknown.
Apply it
Write an evidence table separating visible, user-reported and inferred facts for three sample scenes.
Check your understanding
Your output does not present unsupported inference as an observed fact. Explain the decision and show evidence from your implementation or design. If you cannot demonstrate it yet, revisit the relevant section before continuing.