Some of the most misleading metrics in the current AI hype cycle (aside from popular benchmarks):

- Number of model downloads on Hugging Face (doesn't reflect actual usage)

- Number of tokens processed (doesn't reflect the value and tasks accomplished by those tokens)

- Number of paying subscribers (doesn't reflect exploring users vs users who are actually getting value out of an AI product)

To be clear, I'm not saying that these metrics are not useful, just that they are often misinterpreted as the single indicator that a specific model or AI product is leading its field.