LMCanvas: Object-Oriented Interaction to Personalize Large Language Model-Powered Writing Environments

Beyond their generative capabilities, LLMs demonstrate significant few-shot and Large language models (LLMs) can enhance writing by automating zero-shot performance [4] meaning that they are able to perform or supporting specific tasks in writers' workflows (e.g., paraphrasing, previously unseen tasks with only an instruction and/or a couple creating analogies). Leveraging this capability, a collection of of examples--i.e., a prompt. By leveraging this ability of LLMs, interfaces have been developed that provide LLM-powered tools for writers can potentially automate or augment specific tasks in their specific writing tasks. However, these interfaces provide limited support workflows by using adequate prompts and, thus, further facilitate for writers to create personal tools for their own unique tasks, the writing process. For instance, based only on prompt examples and may not comprehensively fulfill a writer's needs--requiring provided for GPT-3 [17], writers can use LLMs to correct grammar, them to continuously switch between interfaces during writing. In create an outline, produce analogies, or even change the point-ofview this work, we envision LMCanvas, an interface that enables writers of a scene.