Feynman's Letter: Do You Want to Hand-Draw a "Maze Map," or Have AI Generate an "Urban Plan" in One Second? — On Paper Banana
After reading the introduction to Paper Banana (2026.05), a tool quietly going viral in research circles, my feeling is that researchers can finally save the part of their lives spent drawing boxes and connecting lines, and use it to think about the universe instead.
To help you understand why drawing a "system architecture diagram" can drive a PhD student insane, let's talk about diagramming itself.
1. The Status Quo: The Researcher "Fighting Over Pixels" in PowerPoint
Anyone who has written a paper knows that writing text is not hard — the hardest part is drawing that polished-looking "system architecture diagram".
- The pain point: You know exactly how the data flows in your model (from encoder to decoder, with an attention layer in between). But the moment you open Visio or PowerPoint, you become a miserable layout worker: aligning, adjusting colors, drawing arrows... A single diagram takes a whole day, and if you slightly change the algorithm, the whole thing has to be redrawn from scratch. This is the "extremely high physical cost of translating logic into visuals."
- The physical picture (code as graph): Instead of a drag-and-drop canvas, it gives you a text box. You just tell it in human language or pseudocode: "Input A goes through module B (blue), then merges into module D via pipeline C..."
- Automatic topology computation: The tool has a powerful built-in layout engine. It automatically computes the topological relationships between rectangles and arrows, instantly generating high-resolution vector diagrams that meet the standards of top academic venues (like CVPR/NeurIPS). It's like writing a very concise "architectural description," and the machine 3D-prints the villa for you in one second.
2. Paper Banana: The Magician That "Understands Human Language"
Paper Banana's logic is extremely geeky: you shouldn't be operating a mouse — you should be describing logic.
3. The Feynman-Style Judgment: Efficiency Is "Dimensionality-Reduced Expression"
The so-called "system diagram" isn't for showing off.
Its purpose is to collapse that extremely complex, entangled high-dimensional network of logic in your head into a two-dimensional slice that others can understand at a glance.
Paper Banana shows us: anything that can be clearly defined by logic should not be hand-manipulated.
When researchers can escape the torment of formatting and alignment, and generate paper figures simply by describing "how the data flows," this is a perfect dimensional-reduction strike by AI productivity against traditional manual diagramming.
Key takeaway:
When designing complex systems, stop wasting time on alignment guides in drawing tools.
Go find "Declarative" expression tools.
If the logic in your head is rigorous, then there should exist a tool that, like a mirror, instantly maps the wrinkles of your mind onto the most elegant geometric lines in the world.
*Hashtags from the original post: #PaperBanana #AcademicWriting #SystemArchitecture #GenerativeAI #ResearchTools #FeynmanLearning (zhichai research lab)*