Answering Questions Demo:
A learning activity for gaining fluency and confidence when responding to questions from participants
Disclaimer: The content in this demo is not provided by or approved by All of Us.
This demo content was adapted from publicly available materials, but is not maintained or updated to reflect future revisions made by All of Us. For the most current and official All of Us information, please visit joinallofus.org/faq.
Instructions for Demo Testers
Check out the bottom of this page to read How does this work?
⏱️
Please Note: This demo runs on minimal (mostly free) resources.
It may take several seconds to respond to a submission with feedback.
It may take several seconds to respond to a submission with feedback.
Instructions for Learners
A question is randomly chosen from an official FAQ and the learner writes their response in the textbox below the question. After the submit button is clicked, the learner's answer is evaluated and several types of feedback are provided.
Selecting a question...
How Does This Work?
This is a proof-of-concept demo for a learning activity. A question is randomly chosen from an official FAQ and the learner writes their response in the textbox below the question. After the submit button is clicked, the learner's answer is evaluated and several types of feedback are provided:
- Approved Answer: the IRB-approved response to this question on the official FAQ
- Approved Answer Notes: notes taken directly from the official Copy Respository (not all answers have notes)
- Score: Appearing below the learner's answer, the score consists of a percentage and a interpretive. The percentage is based on the similarity score resulting from the Semantic Textual Similarity task in the HuggingFace InferenceAPI.The interpretive label below the score were created for this demo.
- Watch Your Phrasing Alert: regex is used to match strings to the learner's answer. The strings are an unofficial list of words and phrases that the program has asked staff to avoid (the demo list includes only 'research subjects' and 'human subjects'). The alert box appears when a match is found and the matching strings are listed.
- Suiggestions for Improvement: the results of a call to OpenAI API that uses a ChatGPT model to generate a paragraph describing ways the learner's answer could be changed to become semantically closer to the approved answer.