Abstract:Exploring alternative ideas by rewriting text is integral to the writing process. State-of-the-art Large Language Models (LLMs) can simplify writing variation generation. However, current interfaces pose challenges for simultaneous consideration of multiple variations: creating new variations without overwriting text can be difficult, and pasting them sequentially can clutter documents, increasing workload and disrupting writers' flow. To tackle this, we present ABScribe, an interface that supports rapid, yet visually structured, exploration and organization of writing variations in human-AI co-writing tasks. With ABScribe, users can swiftly modify variations using LLM prompts, which are auto-converted into reusable buttons. Variations are stored adjacently within text fields for rapid in-place comparisons using mouse-over interactions on a popup toolbar. Our user study with 12 writers shows that ABScribe significantly reduces task workload (d = 1.20, p < 0.001), enhances user perceptions of the revision process (d = 2.41, p < 0.001) compared to a popular baseline workflow, and provides insights into how writers explore variations using LLMs.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively explore and manage multiple text variants in human - AI collaborative writing tasks. Existing writing interfaces are insufficient in supporting non - linear revision processes. In particular, when considering multiple text variants simultaneously, it is easy to cause document chaos, increase the author's workload, and disrupt the writing process. Moreover, although advanced large - language models (LLMs) can simplify the generation of text variants, the systematic organization, comparison, and modification of these variants remain a challenge. To address these problems, the paper introduces ABScribe - a new writing interface designed to support rapid and structured exploration and organization of multiple text variants in human - AI collaborative writing tasks. ABScribe achieves this goal through the following five interface elements: 1. **Variation Fields**: Allows users to select any part of the text and create interactive variation fields that can accommodate multiple writing variants. New variants can be added without overwriting existing variants. 2. **Popup Toolbar**: Variants are displayed as buttons in a popup toolbar located above the active variation field. Users can view different variants by hovering over the corresponding buttons without disrupting the coherence of the surrounding paragraphs. 3. **Variation Sidebar**: Used in pair with the popup toolbar, the accordion - structured sidebar helps authors view and navigate multiple text variants more easily. 4. **AI Modifiers**: Users can generate new variants by inputting instructions to the AI. These instructions are automatically converted into reusable buttons, and users can apply these buttons to different parts of the document. 5. **AI Drafter**: Users can insert text at any position in the document by prompting the system to call a large - language model. Through these designs, ABScribe aims to reduce task workload, enhance user experience, support context - sensitive variant comparison and revision, and support non - linear and reusable large - language model use. The paper validates the design of ABScribe through a controlled evaluation study. The results show that ABScribe significantly reduces subjective task workload and improves users' perceived quality of the revision process.

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

GhostWriter: Augmenting Collaborative Human-AI Writing Experiences Through Personalization and Agency

Creativity Support in the Age of Large Language Models: An Empirical Study Involving Emerging Writers

Towards Full Authorship with AI: Supporting Revision with AI-Generated Views

"It Felt Like Having a Second Mind": Investigating Human-AI Co-creativity in Prewriting with Large Language Models

Can AI writing be salvaged? Mitigating Idiosyncrasies and Improving Human-AI Alignment in the Writing Process through Edits

Approach Intelligent Writing Assistants Usability with Seven Stages of Action

A publishing infrastructure for Artificial Intelligence (AI)-assisted academic authoring

Shaping Human-AI Collaboration: Varied Scaffolding Levels in Co-writing with Language Models

Human-AI Collaborative Taxonomy Construction: A Case Study in Profession-Specific Writing Assistants

OverleafCopilot: Empowering Academic Writing in Overleaf with Large Language Models

Decoding the End-to-end Writing Trajectory in Scholarly Manuscripts

Let's Get to the Point: LLM-Supported Planning, Drafting, and Revising of Research-Paper Blog Posts

AI-Augmented Brainwriting: Investigating the use of LLMs in group ideation

CoAuthor: Designing a Human-AI Collaborative Writing Dataset for Exploring Language Model Capabilities.

HoLLMwood: Unleashing the Creativity of Large Language Models in Screenwriting via Role Playing

VISAR: A Human-AI Argumentative Writing Assistant with Visual Programming and Rapid Draft Prototyping

LLM-Ref: Enhancing Reference Handling in Technical Writing with Large Language Models

Suggestion Lists vs. Continuous Generation: Interaction Design for Writing with Generative Models on Mobile Devices Affect Text Length, Wording and Perceived Authorship

Inspo: Writing Stories with a Flock of AIs and Humans

Luminate: Structured Generation and Exploration of Design Space with Large Language Models for Human-AI Co-Creation