Create Your Own GPT-3 Chatbot Based on Your Social Media Data 04.21.22 OR 05.26.22
In this workshop, students will learn how to fine-tune the large natural language processing model GPT-3 on their social media data to create a chatbot version of themself. They will then converse with their chatbot-self and compare their own values to those they perceive in the chatbot. In this workshop, we will explore:
- What are our values, and how are these reflected in our social media presence?
- What happens when we create a fine-tuned natural language processing model using such data?
- Do our values also translate?
- Can we truly create chatbot versions of ourselves using this method, or are these representations lacking (and, if so, in what ways?)
This workshop is limited to 30 participants. Cost of participation is free.
Date / Time
The workshop will be held online on 05.26.22 at 10 AM PDT / 7 PM CEST. The workshop will last 1.5 hours.
Participants are asked to complete a number of preparatory steps as homework prior to the workshop. They will also be asked to download their social media data including either Facebook Messenger conversations, Twitter messages, Instagram messages, or WhatsApp messages beforehand, translate into English if necessary, and to format the data for training using the CSV template that will be provided. Each participant will also need to sign up for a free account with OpenAI in advance and install python and the openAI package using pip. Finally, they will need to follow the instructions to fine-tune their model. All preparation instructions can be found here.
Workshop Study (Optional)
This workshop is a part of the study Exploring Embedded Values in Natural Language Processing through Fine-Tuning Foundational Models on Social Media Data. Participation in the study is optional and is not required for participation in the workshop, although priority for enrollment will be given to those who are willing to participate in the study.
- Do I need to know how to program? No. But you will need to use the command line to install certain packages. You should be able to cd into a folder and be comfortable with using the command line. We will show you how to do this in the preparation videos.
- Is this workshop open to students only? No. But because we are running the workshop under the University of Copenhagen and University of Twente we are interested in involving the student body and therefor interested in your affiliation (if any).
- What will I need to do to prepare for this workshop? You will need to 1) sign up for an OpenAI account (free) 2) install Python on your machine 3) install the openAI package with pip 4) download your social media data of your choosing 5) format your data into a CSV of prompts and responses. You can view all the steps at the following page. You should expect to spend two – six hours on workshop preparation “homework” but we promise it will be worth it.
- Who is running the workshop? The workshop is run by Mirabelle Jones (PhD Fellow University of Copenhagen) and Nastasia Griffioen (PhD student at Radboud University).
- What is the study all about? We are running a study in tandem with the workshop about values, social media, and natural language processing. You can consent to be a part of the study when you sign up or you can opt out. The study information sheet is located here. All details will be anonymized.
Sign Ups Are Now Closed
We have reached capacity in the workshop. Thank you to everyone who signed up. We look forward to seeing you on May 26th.