Abstract
An important challenge when using LLMs in business applications is to account for different sources of information that have conflicting values for the same data at the time the LLM performs the inference task based on such data. When multiple potentially conflicting sources of data are available, LLMs used in business applications may need grounding with facts. This disclosure describes a data lineage resolver that can utilize a data lineage catalog to enable LLM output to be verified for grounding based on the data source(s) used for the output and once verified, provided to the business application. The data lineage resolver utilizes the data lineage catalog to check for lineage information, e.g., to determine whether a data source used during training has undergone updates. The data lineage resolver is configured with rules to identify a particular data source that is grounded. The data lineage resolver can be configured to proactively broadcast conflict resolution results to business applications, allowing data stewards or others to update configurations as necessary.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
S, Hari Bhaskar, "Data Lineage Resolver for Grounding Large Language Model Inference", Technical Disclosure Commons, (December 04, 2023)
https://www.tdcommons.org/dpubs_series/6472