Detecting hallucination in a language model

This invention detects when a language model generates made-up information by identifying words in its output that are present in the provided examples but missing from the main content it was supposed to be using. If these "extra" words are uncommon, it's a strong indicator of hallucination. The system can further refine this by calculating frequency scores for suspicious words or by re-prompting the model to check for consistency.

Why it matters: The pervasive integration of LLMs into business operations since 2023 has made reliable hallucination detection a more urgent and valuable capability. While the core problem persists, the ecosystem for deploying and integrating such software has matured.

Status: Under examinationA pending application. It has not been granted, and it has not been abandoned.
How hard to build: SpecializedSoftware logic, LLM integration

3 AI reinvention directions for this patent

See all three

AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.

Reinvent this with AI

Where this one came from

Every morning we go through millions of abandoned patents and pick out 15 worth your time. This was one of today’s.

This one

+10