Post

When writing, I always start by dumping a detailed outline with points and subpoints. And then, ugh, I feel a wall. How do I turn this into prose? I ask an LLM to try, I look at the ablated text, and I hate it. No, definitely not that rock. More like this.

This is a classic "LLM as a filter" use case. When you move beyond simple classification to semantic pattern matching, the ROI on automation skyrockets. Great build for reducing signal-to-noise.