Straight-forwardly true and yet I'd never thought about it like this before. i.e. that there's a perverse incentive for LLM vendors to tune for verbose outputs. We rightly raise eyebrows at the idea of developers being paid per volume of code, but it's the default for LLMs.
I think real endgame is some kind of "no win no fee" arrangement. In the case of the article in the OP, it'd be if they billed clients per "addressed" comment. It's less clear how that would map onto someone selling direct access to an LLM, though.
In the current landscape I think competition handles this neatly, especially since open models have almost the opposite incentive (training on concision is cheaper). As siblings note, Claude tends to be a little less verbose. Though I find they can all be quite concise when instructed (e.g. “just the code, no yapping”).
Straight-forwardly true and yet I'd never thought about it like this before. i.e. that there's a perverse incentive for LLM vendors to tune for verbose outputs. We rightly raise eyebrows at the idea of developers being paid per volume of code, but it's the default for LLMs.