Paper Image

Using open-source AI for educational question answering

Published on:

5 November 2023

Primary Category:

Machine Learning

Paper Authors:

Yann Hicke,

Anmol Agarwal,

Qianou Ma,

Paul Denny

Bullets

Key Details

Uses open-source LLaMA-2 models as a base for educational QA

Applies techniques like retrieval-augmented generation, supervised tuning, and human feedback tuning

Tests on dataset of 10k QA pairs from a CS course's Piazza forum

Modeling techniques together boost answer quality 33% over baseline LLaMA-2

Retrieval-augmented generation has significant positive impact

AI generated summary

Using open-source AI for educational question answering

This paper introduces an innovative solution for scalable and intelligent question answering in education. It leverages open-source large language models to ensure data privacy. The authors use the LLaMA-2 family of models and techniques like retrieval augmented generation, supervised fine-tuning, and human feedback tuning. Experiments on a dataset of 10k QA pairs from a CS course show preliminary evidence that modeling techniques collectively improve answer quality by 33% over baseline LLaMA-2. Retrieval augmented generation is found to be especially impactful.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up