US 12,468,899 B2
Hallucination prevention for natural language insights
Wei Zhang, Great Falls, VA (US); and Victor Soares Bursztyn, Mountain View, CA (US)
Assigned to ADOBE INC., San Jose, CA (US)
Filed by ADOBE INC., San Jose, CA (US)
Filed on May 22, 2023, as Appl. No. 18/321,602.
Claims priority of provisional application 63/500,871, filed on May 8, 2023.
Prior Publication US 2024/0378400 A1, Nov. 14, 2024
Int. Cl. G06F 40/40 (2020.01); G06F 40/186 (2020.01)
CPC G06F 40/40 (2020.01) [G06F 40/186 (2020.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
generating, by a template-based insights engine, a template-based insight with a set of facts from a set of data, the template-based insight being generated in text form based on a text template;
generating, by a language model, a natural language insight from the template-based insight;
before causing display of the natural language insight, determining, by a hallucination gatekeeper engine, a threshold number of facts from the set of facts of the template-based insight missing from the natural language insight by:
searching the natural language insight for each fact in the set of facts of the template-based insight in order of a corresponding string length of each fact; and
based on the search, determining that the threshold number of facts in the set of facts are missing from the natural language insight;
based on the threshold number of facts from the set of facts of the template-based insight missing from the natural language insight, automatically generating, by the language model, a new natural language insight from the template-based insight; and
causing display of the new natural language insight instead of the natural language insight.