Problem solving across 100,633 lines of code in Google Gemini 1.5 Pro.

The code is for generating some animations.

"What controls the animations on the littlest Tokyo demo?"

The model finds the demo and explains the animations are embedded within a glTF model. The videos doesn't explain what glTF is -- apparently it stands for "GL Transmission Format", where "GL" in turn stands for "graphics library" as it does in "OpenGL".

"Show me some code to add a slider to control the speed of the animation. Use that kind of GUI the other demos have."

The show the code and the slider which was added to the scene, which works.

Next, they give it a screenshot of a demo and asked where the code for it was.

There were a couple hundred demos in the system (they never say exactly how many) and it correctly finds the one the matches the image.

"How can I modify the code to make the terrain flatter?"

Gemini finds the function that generates the height and the exact line within the function to modify. It also provided an explanation of why the change worked.

For the last taks they show, they use a 3D text demo that says "three.js".

"How can I change the text to say, 'goldfish' and make the mesh materials look really shiny and metallic?"

Gemini finds the correct demo and showed the precise lines in it to change, along with an explanation of how to change material properties such as metalness and roughness to get a shiny effect.

Problem solving across 100,633 lines of code | Gemini 1.5 Pro demo - Google

#solidstatelife #ai #genai #computervision #llms #multimodal #google #gemini