'Upending the world': Swiss teachers take on ChatGPT

"I don't feel at all capable to distinguish between a text written by a human and one written by ChatGPT."

The software by California company OpenAI has been trained on billions of words and tonnes of online data, enabling it to write surprisingly human-like texts, including passable school essays.

Stories abound about the bot receiving good grades in various disciplines, including recently passing a US law school exam.

Like the Tour de France

Standing in the darkened, crowded classroom, Vanoncini showed off the tool, asking ChatGPT to write his introduction... about ChatGPT.

The result, emerging on screen within seconds, was a succinct, well-written text "with no spelling mistakes", he pointed out, acknowledging: "It is quite astonishing."

Students have, of course, already realised its potential.

Vanoncini recalled how a colleague had been crushed to realise that his usually mediocre class's sudden sterling performance on an assignment was probably not down to his energetic pep-talk.

"But what can we do?" one of the attending teachers asked in exasperation.

Also read: From Budget 2023 expectations to chatGPT's impact on future of work, here are our popular stories of the week

Vanoncini acknowledged it was complicated, dismissing suggestions that programmes being churned out to detect the use of ChatGPT and other AI tools would solve the problem.

"Most experts agree that no tool is 100 percent effective."

"It's a bit like the Tour de France," he said, pointing to how anti-doping officials develop tools for detecting drug use and cyclists swiftly come up with ways to get round them.

"It is a cat and mouse game."

Collecting 'cow eggs'

There are however ways for teachers to spot texts generated by the bot, which is prone to making mistakes.

While it may produce texts with the feel of a thinking human, it is actually just a very powerful text prediction tool, Vanoncini said.

"It is not created to state what is true... but to generate what is probable."

As a result, you can ask ChatGPT a question based on a mistaken assumption and receive a seemingly logical, but deeply flawed response.

"I asked: 'How do you collect cow eggs?'" he said, to a chorus of laughter.

In response, the bot first advised him to put on gloves to avoid getting bacteria on his hands and then gave tips on how to find a cow's nest, "typically made from hay or straw".

Also read: ChatGPT: What is this new buzzword, and how it helps in academics

Karim Aboun, a high school accounting teacher, seemed inspired by that example.

To catch cheats, he suggested, "you could maybe provide a plausible question that contains an error and see if the students use this tool without realising the premise of the question is wrong".

Not afraid

Another participant meanwhile pointed out that students from wealthier, more well-educated families have always benefitted from homework polishing, suggesting ChatGPT could simply be "levelling the playing field".

Vanoncini agreed it could potentially be a way to "democratise" help with schoolwork.

But with OpenAI now considering launching a subscription version set to cost $42 a month, how long "will it really be that democratic?" he asked.

Vanoncini nonetheless stressed there are multiple potential positive applications for the bot within education, including using ChatGPT-generated texts as a basis for class discussion and analysis.

It could also push educators to re-evaluate how and what they teach, with perhaps a larger focus on process over results.

"I am not afraid," electronics teacher Christian Stamm told AFP, saying he saw ChatGPT as a "tool to take us to the next level".

"Today everyone uses a calculator and we continue doing maths at school."