...
By Alice Gibbs
Senior Reporter
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
Online chatbot ChatGPT is seeing a huge boom in popularity and growth. But despite its advanced capabilities, the technology has failed to correctly answer all the questions from a test for 10 and 11-year-olds.
Developed by OpenAI, ChatGPT is an artificial intelligence-powered language model. As AI continues to innovate and evolve almost daily, there are plenty of questions about how it will make our lives better or worse.
Among the excitement and concern about the growth of the technology, many are in awe as it completes incredible tasks, from creating believable photograph-like images to writing well-researched reports.
Read More
Artificial Intelligence: The 21st Century Gold Rush5 min read
Woman Using AI to Redesign Garden Sparks Debate: ‘Cool But Sad’3 min read
How AI Will Make Our Lives Better (And Worse)13 min read
...
But one elementary school test proved too much for the tech. Online learning platform Atom Learning asked ChatGPT to solve the same arithmetic and reasoning questions 5th graders were faced with this month.
Students in the United Kingdom first took the test between May 11 and May 12, with headlines appearing shortly afterward about children coming home in tears because the math test had been so difficult.
When Atom Learning asked ChatGPT to complete the 5th Graders' test questions, the results were shocking.
One of the papers for the 5th Grade students contained 36 questions including addition, multiplication, division and subtraction. Of 36 questions for a total of 40 points, the ChatGPT solved 32 correctly, totaling 34 points, or an 85 percent test score.
2023 KS2 SATs Maths Paper 1
We've shared the answers below so you can test your knowledge, but if you're wondering how you compared to the machines, ChatGPT got questions 9, 21, 24 and 29 wrong.
In response to question 9, ChatGPT incorrectly gave the answer as 4,459, while for question 21, it stated the answer as 4⅓—wrong again.
It listed ⅟₃₂ as the incorrect response for question 24, while it responded with 41 as the answer for 29—another wrong number.
A spokesperson for Atom Learning told Newsweek: "It's interesting and worrying at the same time that 10 to 11-year-olds were handed a test that not even an Artificial Intelligence has managed to complete to its fullest."
Newsweek reached out to OpenAI for comment.
The world has watched in the last year as AI continues to boom, with artificial songs created to sound like huge artists like Drake and a photographer using the technology to beat real photos in a major competition.
Though some people believe that AI technology is set to bring about the decline of human creativity, and even the creator of ChatGPT expressed concern about the potential future of AI, the test proves the technology is not faultless.
Atom Learning was stunned that the tech was unable to complete the children's test questions with a 100 percent success rate and said it made it all the more understandable that many children had felt the test was too difficult.
2023 KS2 SATs Maths Paper 1 - Answers
"While we're aware that AIs such as ChatGPT are not infallible, it's important to remember how these questions were supposed to be tailored for Year 6 (5th Grade) students, which would make anyone suppose that they ought to be 'easy' questions for the likes of adults and computers," said Atom Learning.
"However, in a situation in which not even an AI can find answers to what is supposed to be basic math—it's hard to imagine how young students felt when these same questions were put in front of them on one of the most important days of their lives as school children."
"The Maths Paper 1, Arithmetic, was to be completed in 30 minutes. This means that children had 50 seconds per question," Atom Learning said. "It's now easier to understand the reactions that many of them had."
Meanwhile, anyone who failed the math test above will be pleased to hear ChatGPT performed even worse on another test provided by Atom Learning.
The test, featuring reasoning questions, saw ChatGTP solve 18 of 25 questions, corresponding to 68.6 percent overall.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines