[bg24]

Nice research and structuring into 4-tier layer. For providers like Anthropic and OpenAI, subscription is the entry point for all these, right? Besides the measures proposed in the article, can token usage % determine these clusters of accounts?

[jfim]

There are probably various metrics like language used to prompt the model, number of hours per day spent prompting, and many others.

There are quite a few other mitigations that could be done by providers that aren't mentioned in the article.