公开(公告)号：US20250124235A1

公开(公告)日：2025-04-17

申请号：US18485204

申请日：2023-10-11

Applicant: ADOBE INC.

Inventor： Victor Soares BURSZTYN ,  Xiang CHEN ,  Vaishnavi MUPPALA ,  Uttaran BHATTACHARYA ,  Tong YU ,  Saayan MITRA ,  Ryan ROSSI ,  Manas GARG ,  Kenneth George RUSSELL ,  Eunyee KOH ,  Alexandru Ionut HODOROGEA

IPC: G06F40/40 ,  G06F40/279

Abstract: Methods and systems are provided for using generative artificial intelligence to evaluate fine-tuned language models. In embodiments described herein, natural language text snippets are generated via a generative language model based on corresponding data. A language model is fine-tuned into a fine-tuned language model via a language model fine-tuning component using the natural language text snippets and the corresponding data as training data. Independent natural language text snippets are generated via the generative language model based on the corresponding data. Each independent natural language text snippet is different than each corresponding natural language text snippet. An evaluation metric of the fine-tuned language model is generated via an evaluation component based on the independent natural language text snippets and the corresponding data.

公开(公告)号：US20240378400A1

公开(公告)日：2024-11-14

申请号：US18321602

申请日：2023-05-22

Applicant: ADOBE INC.

Inventor： Wei ZHANG ,  Victor Soares BURSZTYN

IPC: G06F40/40 ,  G06F40/186

Abstract: Methods and systems are provided for hallucination prevention for natural language insights. In embodiments described herein, a template-based insight with a set of facts is generated by a template-based insights engine. The set of facts are generated from a set of data and the template-based insight is generated based on a text template. A natural language insight is generated from the template-based insight through a language model. If a threshold number of facts of the template-based insight are missing from the natural language insight as determined by a hallucination gatekeeper engine then a new natural language insight is generated from the template-based insight through the language model.