Researchers from the University of Illinois tested the free version of ChatGPT to see how well it could do in an undergraduate aerospace engineering class, and it was found that even though ChatGPT scored an A in basic math questions, it couldn't do well on open-ended, complex questions and scored a D on them. Overall, ChatGPT scored 82% and got a B, while the scores of average students were a bit higher at 84.85%.
These findings show that students can easily pass the course with a B by relying on ChatGPT, but it can only be through getting math questions right and not doing well on the questions related to critical thinking. This means that they could pass, but they wouldn't learn anything by relying completely on ChatGPT. The professor who is leading the course, Melkior Ornik, says that ChatGPT acts as a calculator, so he's planning to add more open-ended and project-based questions in the course so the students can develop thinking skills.
The researchers found that ChatGPT makes incorrect answers, even if it has all the course material, like it once invented the term ‘quasi-periodic oscillations’ which was never once mentioned in the class. The researchers also said that this study was mainly focused on students who do not put in the effort and go to ChatGPT for their coursework. That's the reason the researchers used a free version of ChatGPT because many students wouldn't pay for the upgraded model. ChatGPT acted just like the student and did its homework like others did.
When the researchers told ChatGPT about its incorrect answers, it adjusted to a better answer, but there was not much improvement. If ChatGPT scored 90% on homework earlier in the course, it improved to around 92% by the end of the semester. This shows that even though ChatGPT tried to learn, it wasn't able to show any real progress.
Image: DIW-Aigen
Read next:
• Big Tech Giants Google, Microsoft, Apple, and Meta Drained 132 Million Cubic Meters of Water, Filling 52,938 Olympic Pools!
• Cybercrime Losses Reach $16.6B in 2024, Seniors Lose $4.8B, FBI Gets 860K Complaints, Crypto Tops $9.3B
These findings show that students can easily pass the course with a B by relying on ChatGPT, but it can only be through getting math questions right and not doing well on the questions related to critical thinking. This means that they could pass, but they wouldn't learn anything by relying completely on ChatGPT. The professor who is leading the course, Melkior Ornik, says that ChatGPT acts as a calculator, so he's planning to add more open-ended and project-based questions in the course so the students can develop thinking skills.
The researchers found that ChatGPT makes incorrect answers, even if it has all the course material, like it once invented the term ‘quasi-periodic oscillations’ which was never once mentioned in the class. The researchers also said that this study was mainly focused on students who do not put in the effort and go to ChatGPT for their coursework. That's the reason the researchers used a free version of ChatGPT because many students wouldn't pay for the upgraded model. ChatGPT acted just like the student and did its homework like others did.
When the researchers told ChatGPT about its incorrect answers, it adjusted to a better answer, but there was not much improvement. If ChatGPT scored 90% on homework earlier in the course, it improved to around 92% by the end of the semester. This shows that even though ChatGPT tried to learn, it wasn't able to show any real progress.
Read next:
• Big Tech Giants Google, Microsoft, Apple, and Meta Drained 132 Million Cubic Meters of Water, Filling 52,938 Olympic Pools!
• Cybercrime Losses Reach $16.6B in 2024, Seniors Lose $4.8B, FBI Gets 860K Complaints, Crypto Tops $9.3B
