GPT-6 Luna, Haiku 5.5 and Gemini 3.8 Flash compared on the same physics test
A soft rod slides down a ramp and must land in a glass tube while a stiffness slider drops from 100% to 10%, showing how GPT-6 Luna, Haiku 5.5 and Gemini 3.8 Flash each simulate bending and landing.
🚨 Same physics test. Three AI models. One slider that breaks everything. Watch what happens as stiffness drops from 100% to 10% the setup: a glowing soft rod slides down a ramp and has to land in a glass tube the models: → GPT-6 Luna → Haiku 5.5 → Gemini 3.8 Flash at 100% stiffness the rod behaves like a solid stick at 75% it starts to bend at 50% it folds on the ramp at 25% it flops at 10% it turns into a noodle that has to find its way into the tube every model has to handle the same slider. each one decides on its own how bending, falling and landing should look physics is where models quietly show whether they model the world or only imitate it which one handled the noodle best? 👀
This post quotes another post: x.com/romance0x/status/2108123784880210371