Project vision

Learn by building a mathematical world

GPUSims explores how AI, executable simulation, numerical observation, and visual explanation can make difficult mathematics and physics more approachable.

Motivation

From equations to observable behavior

Advanced physical theories are often introduced through dense notation and long derivations. A simulation adds another path: change a parameter, observe the response, inspect the numerical state, and then return to the equations with a concrete phenomenon in mind.

GPUSims does not remove the need for mathematics. It creates an interactive bridge between symbolic ideas, numerical algorithms, GPU implementation, and visible behavior.

Visual intuition

Fields, waves, particles, surfaces, and phase changes become objects that can be manipulated rather than diagrams that remain fixed on a page.

Progressive explanation

AI can write an initial article and then answer increasingly focused questions about any equation, approximation, or code path.

Human–AI collaboration

A structured conversation with executable feedback

Ordinary code generation ends when text is produced. GPUSims continues the process by compiling the generated code and producing evidence: errors, numerical values, and images.

  1. Hypothesis
  2. Execution
  3. Observation
  4. Correction

The user can question each stage. Was the schema appropriate? Did the shader implement the intended equations? Are conserved quantities stable? Does the rendered geometry match the model?

Research question

Can AI learn through simulation?

Large language models learn powerful patterns from text, but a simulation offers a different kind of loop. An AI can propose a model, execute it, observe numerical and visual consequences, revise the implementation, and repeat the experiment.

This is a research direction, not a completed claim. GPUSims currently provides a practical environment for experimentation. Whether repeated simulation feedback produces a more grounded physical world model remains an open question.

Scope

Accessible experimentation, not automatic truth

Beautiful output can still be wrong. AI explanations can contain mistakes. WebGPU generally uses single-precision arithmetic, and educational simulations may omit important physical effects.

The project therefore emphasizes inspection and questioning. Users should compare results with theory, known limiting cases, conservation laws, independent calculations, and expert review when accuracy matters.

Long-term direction

A browser-based laboratory for more people

The goal is to let learners, teachers, programmers, and researchers begin with a question and quickly construct an inspectable mathematical experiment. GPUSims is intended as a doorway into deeper study—not a replacement for it.

Open App Read the Manual →