Use AI Hallucinations to Your Advantage
Imagine you’re writing a piece of fiction and you get stuck developing the world. You have descriptions of a few different aspects of it, but a particular gap continues to feel just out of reach. You decide to use an AI to help fill the gap.
Enter: hallucinations as a feature.
Because LLMs reflect a linguistic model of the world, they sometimes produce output that lacks a sensible semantic understanding. Glue on pizza, eating rocks, and smoking a few cigarettes a day during pregnancy are some recent examples of this phenomenon, usually termed a “hallucination”.
If you’re trying to use the AI to solve real-world problems, hallucinations are a significant drawback. The unpredictability and the misinformation, even when recognized as such, add unnecessary friction.
However, if used for creative work or work where there is no “right” answer, hallucinations might not be a bug. They can be a feature that helps you imagine what an alternate universe — such as the one from that novel — might look like. This is one the ways to use AI “to brainstorm”, made practical. The AI can expand the consideration of possibilities in ways we may be less likely to consider precisely because it’s not constrained by sense-making the way humans are.
