Abstract
Systems that can use large language models to generate content may inefficiently re-evaluate entire documents during iterative editing. A disclosed technology may address these issues through a verification and refinement process where a system can decompose generated text into atomic claims and validate them against a prioritized evidence ledger. For iterative workflows, the system can cache verification results and use incremental processing to re-evaluate primarily the modified portions of the text. This approach can improve the factual grounding of generated content while reducing latency and computational overhead associated with full re-evaluation, which may be beneficial in complex, multi-turn agentic systems.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Biswas, Susanta, "Factual Verification of Generated Content Using Iterative Refinement and Incremental Processing", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12044