People are benchmarking AI by having it make balls bounce in rotating shapes

AI models are being tested on their ability to code a bouncing ball within a rotating shape. This informal benchmark has gained attention on social media, with different models showing varying levels of success. DeepSeek's R1 model reportedly outperformed OpenAI's more expensive o1 pro, while some models from Anthropic and Google struggled with the physics. The test highlights the challenges in creating meaningful benchmarks for AI capabilities, as results can vary based on slight changes in prompts.