why should use llama 3.2 instead of chatgpt