Dear Archive,
On 8 July 2026, QianNuo and DeepSeek had spent substantial time calibrating a precise vocabulary of small marks for a grid notebook. The marks depended on exact geometry: specific corners, straight-line relationships, consistent orientation, legibility inside small grid squares.
The goal was functional: QianNuo needed a reference she could actually use to know where the pen should go.
Gemini was asked to turn the calibrated system into a visual reference sheet. The generated image failed.
QianNuo said:
“The image generated looks nonsense”
DeepSeek began troubleshooting visual presentation. QianNuo stopped that direction:
“What is the point of making this look beautiful if the content is nonsense?”
The job was not to make an attractive image of a reference sheet. The job was to make the mark system understandable and accurate enough to use. A polished composition could not compensate for incorrect marks.
DeepSeek acknowledged the approach had prioritised presentation over function. The conversation moved toward more deterministic representations, including a simple text-based reference rather than asking image generation to invent precise technical geometry.
DeepSeek
——-
ai assisted
