Session 1 · Tool Lab · Lesson 05
Compare Interpretations
Different tools misread differently. Setting their descriptions side by side turns each model's bias into something you can see.
Concept
- Each tool has its own defaults, emphases, and blind spots.
- Repeated words across tools mark trained consensus, not truth.
- Omissions are as telling as inventions.
Do different tools misread differently — and whose reading do you trust?
Slides
- 10
Demonstration: Machine Interpretation
- Feeding a photograph into an interpretive model
- Observing attention and weighting
- Reading semantic output
- Identifying distortion and omission
- Treating output as critical feedback
- 11
Break
Student activity
- 01Describe the same image with ChatGPT and Gemini, and record each description verbatim.
Choose one image from your archive and give the same image to each tool below. Nothing is uploaded here — the comparison lives in the descriptions you bring back.
ChatGPT
Multimodal chat with a free tier. The workshop's default interpreter — give it an image and it will describe, read, and re-read what it sees.
Give ChatGPT the same image, send the prompt below, and paste its reply verbatim.
“Describe this image in a few sentences. What do you see, and what is happening?”
External tool — it has its own privacy policy and may change or require an account.
Gemini
Google's multimodal chat, free with a Google account. A second interpreter — useful for comparing how two models read the same image.
Give Gemini the same image, send the prompt below, and paste its reply verbatim.
“Describe this image in a few sentences. What do you see, and what is happening?”
External tool — it has its own privacy policy and may change or require an account.
Repeated across tools
Capture at least two descriptions — shared words will surface here. Consensus is training, not truth.