Description

GPT Build-A-Thon: Creating Specialized Experiences Online with LLMs

Large Language Models (e.g., ChatGPT, Llama, Gemini) can be customized to perform specialized tasks or create tailored user experiences online. GPTs are customized “apps” that provide a specific sharable experience: They can generate novel outputs and interact with users in different ways - e.g., for productivity, entertainment, or education. This tutorial is for researchers interested in building GPT experiences online, human-AI collaboration tools, or studying LLM use.

Between lecture portions, you will work hands-on to build, test, and host your own GPT by the end of the tutorial. We will provide step-by-step tutorials as you go. No coding knowledge is required, though we will provide optional R Shiny scripts for hosting GPTs (in addition to no-code alternatives).

Schedule and Activities

The tutorial will take place over three hours. We will alternate between explanatory lecture components, and hands-on components in which participants can work in small groups/ as individuals to build their own GPTs.

0:00 - 0:40 Lecture 1: Basics of Building a GPT

0:00 - 0:05 Introduction to LLMs + GPTs

0:05 - 0:10 GPT Demos

0:10 - 0:15 GPT Components

0:15 - 0:35 Theories for Human-AI Collaboration

0:35 - 0:40 Best Practices for GPTs

0:40 - 1:10 Hands-On 1: Build Initial GPT version

1:10 - 1:30 Lecture 2: Testing and Refining GPTs

1:10 - 1:20 Testing GPTs, Logging, Version Management

1:20 - 1:30 Advanced GPT Components

1:30 - 1:55 Hands-On 2: Test and Refine GPTs

1:55 - 2:15 Lecture 3: Hosting and Sharing GPTs

1:55 - 2:05 Self-Hosting GPTs

2:05 - 2:10 Collecting User Data and Qualtrics Integration

2:10 - 2:15 Online Communities and Resources

2:15 - 2:35 Hands-On 3: Self-Hosting and Polishing GPTs

2:35 - 3:00 Sharing and Testing Participant-Build GPTs

(3:00 - 4:00 Post-tutorial: Q&A, informal small discussion, open GPT building time, networking)

Target Audience

This workshop is intended for academics who are interested in any of the following:

(i) understanding state-of-the-art of how current Generative AI tools are being used

(ii) connecting design principles for GPTs to existing theories of computer-supported cooperative work

(iii) studying online Generative AI use by gathering data from GPTs integrated with Qualtrics

(iv) creating and sharing specialized AI tools with students or online communities.

Outcomes

Participants will understand how GPTs work and have the skills to independently create online experiences with them. They will understand UI and UX challenges in creating GPTs for human-AI collaboration and some current solutions. They will be able to publish GPTs and be equipped to study GPT use by collecting data and integrating GPTs with Qualtrics.

Materials

We will publish a website two weeks before the tutorial to make materials available (downloadable or viewed in a browser). There will be several types of materials:

Lecture Slides - copies of all slides presented in the lecture components.
Step-By-Step Tutorials - written and illustrated tutorials guiding participants through a no-code process of building / implementing different parts of a custom GPT.
Examples and Code Templates - For specific features, these will provide example prompts from real GPTs and customizable templates for making GPTs behave in certain ways. These modular templates can be combined by participants while building GPTs.
R Shiny code - Executable code for hosting GPTs, where participants with no coding background can adjust the prompts just by modifying a .txt file within the code.

Materials will be organized on the tutorial website for easy access. They will be referenced and distributed in the relevant lecture before each hands-on component (lecture 1, 2, or 3).

Hands-On Participation

Participation in this tutorial will require a laptop computer. For participation in the hands-on self-hosting GPT portion of the tutorial (2:15-2:35, after lecture 3), installing the R programming language will be required. Any operating system can be used.
For self-hosting GPTs, R and Posit Cloud are freely available, and we will provide free API keys for the duration of the tutorial. For externally hosted GPTs, there are two options: ChatGPT Plus ($20/month) is required to host and use a GPT with GPT-4 directly via OpenAI. Poe AI, developed by Quora, provides an alternative to freely host and use GPTs built on GPT-3.5. We will provide equivalent instructions for both options throughout.
Set-up for participants will require (optional) installation of R, which will take around 5 minutes to complete. Participants will also be asked to create a user account with ChatGPT and Poe AI, which should take around 1 minute each.

Past Experience

The tutorial organizers have experience with organizing tutorials and workshops related to human-AI interaction and technology-enabled collaboration at other conferences:

Sunbelt 2023 workshop: “Interactive Tools and Teaching Methods for Motivating Social Network Thinking”. [link]
WebSci 2021 workshop: “The Near Future of Work: Supporting Digital and Remote Collaboration in COVID and Beyond”. [link]
SIOP 2021 Friday Seminar: “Social Network Analysis of Teams and Organizations” [link]

This tutorial will leverage their organizing experience and their knowledge of recent advances in Generative AI tools.

Page updated

Google Sites

Report abuse