What is ChatGPT? Concerns from language teachers
Recap of Tech Talk for Teachers #4 (12 Dec 2022)
ChatGPT in 3 words: prototype AI chatbot. We have seem fellow countrymen from Singapore (and across the world) lax lyrical about what it can do. Within 5 days of launch, there were over a million users but what exactly is its significance in terms of teaching and learning?
Apart from writing poems, what else can it do and why does it matter?
With inputs from teachers from Zhong Hua Secondary School, Hougang Secondary School and more, the session explored what ChatGPT can do.
- Many articles online expressed concern about how ChatGPT was being used in college admissions essays, and by extension:
- All literary ‘creative works’ could essentially be “plagiarised”, “threatening” all humanities-type assessments which have long form essays and can accept a digital submission (e.g. English/ General Paper (GP) Essays, Poems and more)
- For computing teachers: it could write code (to a good degree of accuracy) even though it was banned by Stack Overflow, a Q&A site used by coders
During the session, we noted how its ability to write is particularly insidious as ChatGPT is capable of writing like a particular level or grade. For instance, you can ask ChatGPT to write like a primary school children or adopt more academic language. Video demo here.
In the session, we discussed 3 other instances in which ChatGPT might be problematic:
- It is confidently and hilariously wrong in defending why computing on an abacus is superior to CPU computation. See Andrew Ng (Stanford Professor/ Founder of Coursera)'s demo here
- Bad diet advice - when asked, "should I eat an avocado when there's a lot of uric acid production in my gut", ChatGPT confidently responded yes with references that are entirely made up. Should teachers then also check the integrity of sources cited?
- Even though safeguards exist to hide dangerous content, it seems like there are tacky workarounds that pose risk to users. For instance, there was a Twitter thread showing how a user managed to convince ChatGPT to teach it how to make a molotov cocktail
A quick note about the tech behind what makes ChatGPT so powerful:
The Language Model used in ChatGPT is trained using Reinforcement Learning from Human Feedback (RLHF). More information here
We had good attendance from language teachers. While not ChatGPT itself, on an related note, we saw an interesting applications of AI suggested by a fellow teacher Joyce. She used Quillbot to teach grammar:
Joyce explained how she asked students to type paragraphs and review Quillbot's writing suggestions. In this instance, students had to actively review whether the suggestions are legitimate. Teachers then can facilitate and prompt students to think critically about whether the suggestions make sense.
In particular, Joyce noted that since Quillbot (in its current state of tech) is seldom able to decide choice of words in context, this was a uniquely important insight that students could pay attention to when deciding whether to accept suggested rephrase. The machine may not be able to guess your intention of the paragraph. The bot is not always right.
A well trained machine can proofread for correctness where there are clear language rules but cannot give you an absolutely right answer. In the next session/ post, we will explore the limits of GPT detector and originality.AI.
Where to try/ what else to try?
The author strongly encourages you to give the following tools a shot as the best way to understand the potential of AI is to try it for yourself!(: