The language model is allowed to use:

• selenium webdriver running in non-headless mode on macOS or linux

• 100 loops/agent completions per attempt

• 1x Nvidia A100 GPU

Rules

• I'm allowed to write code, finetune ai models, run scripts

• I'm NOT allowed to help the model during its attempts

• Win condition is the "order placed" screen

Pizza chain of choice

• Probably dominoes or something

If a good enough multimodal model doesn't exist, I will attempt to make it before the deadline