ChatWise

A personal knowledge management (PKM) tool that Introduces large language models to the whole PKM process and revolutionizes the efficiency of knowledge-acquiring.

Designed by Diane Hu
Timeline 2023 March (4 months)
Tools
Figma
ChatWise running on a laptop in a café — a concept map on the left, an annotated conversation on the right
Fanned-out screens of the ChatWise interface — an annotated conversation flanked by summary, metadata and note panels

Background

ChatGPT sprinted to 100 million monthly active users in 2 months; OpenAI investment thesis map of generative AI applications GPT-4 exam results ordered by GPT-3.5 performance across academic tests News coverage of school districts blocking ChatGPT amid fears of cheating, and OpenAI's work on mitigations
How can we make responsible usage of AI tools in academia?

Research

Learning vs. copying: the knowledge loop of connect, organize, develop, express, note and encounter What is knowledge and how to make personal knowledge management — Peter Drucker, Niklas Luhmann and Tiago Forte How can knowledge be classified — by analytic philosophy and by source of knowledge Why ChatGPT is a powerful tool to acquire knowledge, with the features of ChatGPT set against the principles of PKM

Interviews

The interview aims to investigate user experiences of utilizing ChatGPT primarily as a learning resource, as opposed to a task-completion tool.

Four interviewees and the interview outline: background, ChatGPT use, knowledge acquiring and knowledge management

Key Findings

Annotated ChatGPT session showing overload of text, lack of structure, hard to revisit, piled-up history and cost of switching between apps

Experiments

Experimental subject, purpose and content: comparing PKM with and without ChatGPT Experimental tools — capture, note-taking, mind-mapping and all-in-one workspace apps
Task 1 flow, input and output, with findings Task 2 flow, input and output, with findings Task 3 flow, input and output, with findings Task 4 flow, input and output, with findings

Affinity Diagram

Affinity diagram clustering findings and user quotes under encounter, note, connect and organize

Pain Points

Pain points for encounter, note, connect and organize in the knowledge loop

Persona

Two personas — a 35-year-old content creator and a 25-year-old student researcher — with their AI engagement, PKM strategy and struggles

Design Concept

Five-step pyramid from copying LLM results up to introducing LLMs to the whole PKM process, marked we are here

Analogy

ChatGPT and podcast — using analogy to view existing issues through the lens of another domain Why podcast: conversational, unstructured text, knowledge extraction Five practices of information management borrowed from podcasts — quick positioning, comprehension, retrieval, organization and key insights

Competitive Analysis

Glasp, Sensecape and Flux compared on product positioning and core features
Opportunities: the market focuses on content generation, leaving content organization and digestion with LLMs untapped

Design Objectives

Six design objectives, from amplifying knowledge acquisition to optimizing user experience

Core Functions

Explore mode and digest mode around the knowledge acquisition loop Concept map features for explore mode and annotation toolkit features for digest mode

Features

Three features: non-linear concept mapping, effortless retrieval and powerful annotation

Information Architecture

Information architecture from the homepage down through explore mode and digest mode

Wireframe Ideation

Concept graph ideation — force-directed graph, card graph and logic chat, each with advantages and disadvantages
Explore mode vs. digest mode, option 1: combinable tabs Explore mode vs. digest mode, option 2: versatile notes Explore mode vs. digest mode, option 3: comprehensive concept map

Final Design

Switch between explore mode and digest mode Clicking the preview of the concept map switches the workspace between the two modes
Explore view — the tri-fold mapping framework of linear, associative and sequential connections, with follow-up questions and response re-editing Zooming into the concept graph to reach a single conversation, with the legend of links, notes, highlights and instructions Digest view — annotation tools alongside the conversation, and the ChatWise visual style guide

Storyboard

Six-panel storyboard: doubts, capturing insights, weaving connections, structuring a library, new connections, and sharing with partners