Writing gadgets have been extensively investigated for many different writing goals and activities. The focus of recent developments in writing assistants has been on Large Language Models (LLMs), which enable individuals to produce material in response to a prompt by providing its purpose. Significant developments in LLMs such as ChatGPT and their use in popular products highlight their potential as writing assistants. However, the human-computer interface with these assistants reveals significant usability issues, including coherence and fluency in model output, trustworthiness, ownership of generated materials, and predictability of model performance.
While some of the interactive components of writing assistants have been studied in previous publications, there is still a focused attempt to meet overall writing goals and approach their interactions from a usability perspective. These issues often result in users needing assistance in using the tools successfully to achieve their writing goals and sometimes cause users to give up altogether. Researchers from McGill University and the University of Montreal are examining interface design for LLM-powered intelligent writing assistants, focusing on human activities and drawing influence from previous research and design literature. They also propose using Norman’s Seven Business Stages as a design model for LLM-powered intelligent writing assistants and analyzing usability implications.
The cyclic cognitive model known as Norman’s Seven Stages of Action is frequently used to understand users’ thought processes and associated physical activities. It is mainly used to report the system interface design. The seven action steps are (a) goal development, (b) plan, (c) identification, (d) performance, (e) realization, (f) interpretation, and (g) comparison, as illustrated in Figure 1. Planning and making up the phases Determining and implementing them is the stage of implementing the interaction, and the evaluation stage consists of realizing, interpreting and comparing the stages. User interactions are based on a mental model of the system that they have developed from previous assumptions. They assert that this model enables the creation and evaluation of interfaces that facilitate precise interactions with LLMs at different stages.
🚀 Join the fastest ML Subreddit community
They suggest that an effective LLM-based written help should answer questions relevant to the different stages of design communication and give the user key skills. They give an example that was heavily influenced by their initial efforts to use OpenAI’s Codex to write tutorials to further illustrate their point. In a typical interaction, the user starts by setting a primary goal, such as creating a lesson on how to use matplotlib to plot data points. Then they break down the goal into manageable components to help them decide how to approach the writing assistant.
The main objective can be divided, for example, into three sub-objectives:
- Composing sections of the tutorial
- Provide appropriate instructions for installing the library in different contexts
- Produce and explain code snippets
- Increase the readability of the tutorial
Although it has a narrower scope and can come after several cycles of the framework, each step in this situation can also be regarded as a sub-goal. When clients ask for help from a writing assistant, they often describe their request and then complete it via the interface, for example, “Write a code snippet to plot a scatter plot using matplotlib given data points in a Python list and provide an explanation of the code.”
The Performance phase can have different interface capabilities for changing and updating prompts, while the Select phase may contain systems for recommending alternative prompts for the form. The implementation phase is influenced by the users’ previous conceptual models, their jobs, their domain experiences, and both. When the writing assistant produces an output, the user reads, understands, and adjusts pre-existing mental models after their knowledge and skill. For example, a user with a lot of experience with matplotlib might be better able to detect any unexpected textures or errors in the resulting code. In addition, it may be required to run any existing unit tests or execute the code snippet produced in the IDE to compare results with resources in other contexts.
They contend that applying Norman’s seven stages of action as a model for investigating user behavior using LLM-based writing tools can provide a useful basis for investigating and designing subtle interactions during the goal formulation, implementation, and evaluation phases. It is possible to identify important interactions and direct the design of the writing assistant to aid the work of creating tutorials by asking questions relevant to each step. It is possible to solve specific usability problems in the design of LLM-based writing tools by analyzing the devices and their features across interaction design dimensions defined by the framework. More ambitiously, they point to understudied areas of study in human-to-human LLM interactions, such as compatibility with user preferences, design of effective stimuli, and the explainability and interpretability of model output.
scan the paper. Don’t forget to join 20k+ML Sub RedditAnd discord channelAnd Email newsletter, where we share the latest AI research news, cool AI projects, and more. If you have any questions regarding the above article or if we’ve missed anything, feel free to email us at Asif@marktechpost.com
🚀 Check out 100’s AI Tools in the AI Tools Club
Anish Teeku is a Consultant Trainee at MarktechPost. He is currently pursuing his undergraduate studies in Data Science and Artificial Intelligence from the Indian Institute of Technology (IIT), Bhilai. He spends most of his time working on projects aimed at harnessing the power of machine learning. His research interest is in image processing and he is passionate about building solutions around it. Likes to communicate with people and collaborate on interesting projects.