Wharton's Ethan Mollick finds GLM-5.2 Deep Think Max introduced code errors where older GPT-5.2 successfully rendered WebGL shaders · Digg