Why ChatGPT Misrepresents Experts When Source Data Is Sparse
April Dunford explains that generative AI can confidently invent an expert's position when little relevant source material exists. Instead of recognizing the gap, it combines generic subject knowledge with assumptions about what the expert probably believes.
- Sparse source data encourages plausible fabrication
- Repeated public commentary may still be insufficient context
- Fluent answers can conceal incorrect attribution
- Large, disparate datasets are better suited to AI synthesis
“Here's April. She's a positioning expert. So she probably thinks this. And it turns out it's, you know, it's wrong. It's really wrong.”
“Sparse data, like that's a problem.”