What's the benefit of OpenAI charging per-token instead of per-character or per-word?
Since token algorithms change model-to-model and version-to-version, it seems like they've added a lot of complication for no actual benefit to the user except for a little peek under the hood.
Is there a benefit to this scheme that I'm not seeing? Is there some way to game the system otherwise?
Since token algorithms change model-to-model and version-to-version, it seems like they've added a lot of complication for no actual benefit to the user except for a little peek under the hood.
Is there a benefit to this scheme that I'm not seeing? Is there some way to game the system otherwise?