LLMs increases coverage not efficiency

LLMs likely don't help you complete tasks that much more efficiently. There's already plenty of work done about work slop and whether what people are producing with LLMs is better and it largely seems like it isn't. Instead, LLMs allow you to try a lot more things a lot more rapidly. That in itself is measurable. I also think that there will be two major byproducts from this state of affairs, if it's true. One is that people will try many more things before deciding on one approach, when assisted by an LLM, resulting in more sensitivity to final form than previously. If you can have an LLM give you a hundred different titles for your paper, you will play around more (this is in direct opposition to the homogenization thesis). 2. is that the lower bound of what people will do will rise in many areas where an LLM could do something at least as good as a human used to be able to on its own and therefore or to seem unique, you'll have to try a lot harder.