Skip to main content Skip to navigation

ABScribe: Rapid Exploration & Organization of Multiple Writing Variations in Human-AI Co-Writing Tasks using Large Language Models

Project Overview

The document discusses the innovative use of generative AI in education, focusing on ABScribe, a cutting-edge writing interface that incorporates Large Language Models (LLMs) to enhance the writing experience. ABScribe facilitates a more effective writing process by allowing users to explore and organize multiple writing variations simultaneously, which alleviates cognitive overload and improves satisfaction with the revision process compared to traditional linear editing methods. Its key features include Variation Fields for storing different writing options, a Popup Toolbar for easy comparison of variations, and AI Modifiers that enable the creation of reusable prompts. A user study highlighted the significant improvements in task efficiency and overall user satisfaction, underscoring the effectiveness of non-linear, parallel exploration in writing. This application of generative AI not only streamlines the writing process but also fosters a more engaging and productive learning environment, demonstrating the transformative potential of AI technologies in education.

Key Applications

ABScribe - a writing interface for exploring and organizing writing variations using LLMs

Context: Educational context for writers, specifically in tasks that require revision and exploration of text variations. Target audience includes students, researchers, and professionals engaging in writing.

Implementation: The interface was designed through an iterative process involving low to high-fidelity prototypes, and validated through a user study with 12 participants comparing it to a baseline AI-integrated editor.

Outcomes: The user study indicated that ABScribe significantly reduces task workload and enhances perceptions of the revision process, allowing for a more efficient exploration of multiple writing variations.

Challenges: Some users preferred the linear representation of variations seen in traditional interfaces and expressed concerns about the unfamiliarity of non-linear workflows.

Implementation Barriers

User Experience Barrier

Users accustomed to linear editing interfaces might struggle with the non-linear exploration of variations, leading to potential confusion and resistance to adopting the new tool.

Proposed Solutions: Providing training or tutorials to help users transition from linear to non-linear workflows could facilitate adoption.

Cognitive Load Barrier

Managing multiple variations simultaneously may lead to cognitive overload for some users, especially those unfamiliar with such workflows.

Proposed Solutions: Implementing visual cues or simplified interaction methods could reduce cognitive load, making it easier for users to navigate and manage variations.

Project Team

Mohi Reza

Researcher

Nathan Laundry

Researcher

Ilya Musabirov

Researcher

Peter Dushniku

Researcher

Zhi Yuan "Michael" Yu

Researcher

Kashish Mittal

Researcher

Tovi Grossman

Researcher

Michael Liut

Researcher

Anastasia Kuzminykh

Researcher

Joseph Jay Williams

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Mohi Reza, Nathan Laundry, Ilya Musabirov, Peter Dushniku, Zhi Yuan "Michael" Yu, Kashish Mittal, Tovi Grossman, Michael Liut, Anastasia Kuzminykh, Joseph Jay Williams

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies