FarWay

@webgl_webgpu @cerebras Thanks! every cycle the Critic AI looks at the rendered frame and scores it on stuff like nebula density / star field richness. it flags the weakest part, then the Composer AI generates the next shader to fix that, using the last few rounds of code+scores to build on what worked

@FarWay1104 @cerebras This is super cool! Love the render-feedback loop and not just AI writing GLSL. Curious on how are you scoring whether a generated shader is "good"?

Built for Cerebras x Gemma 4 hackathon: Neural Shader JIT Compiler. An autonomous agent loop that critiques and rewrites GLSL code based on real-time rendering feedback. Pairing Gemma 4 with @Cerebras inference cuts shader iteration time from minutes to <6s. Live art, evolved.

@cerebras github.com/SrijanSuresh/N…