Reduces hallucinations by teaching models 'epistemological humility'—the ability to admit they don't know something—using synthetic non-existent terms.
arXiv · March 19, 2026 · 2603.17504
The Takeaway
Instead of complex RLHF pipelines, this approach uses a targeted SFT dataset of 'hypothetical terms' to fix the model's tendency to guess. It significantly improves factuality scores while maintaining general intelligence, offering a simpler path to reliable LLM deployments.
From the abstract
Large language models (LLMs) often hallucinate, producing fluent but false information, partly because supervised fine-tuning (SFT) implicitly rewards always responding. We introduce $\textit{HypoTermInstruct}$, an SFT dataset (31,487 responses for 11,151 questions) designed to teach models epistemological humility-the ability to recognize the limits of their own knowledge and admit uncertainty. This is achieved through questions about non-existent "hypothetical" terms. We also release $\textit{