The research was undertaken by the American educational website EduRef, which decided to compare the skills of the generator and fresh university graduates. For this purpose, a group of lecturers was hired who prepared guidelines and suggestions for essay writing on US history, research methods, creative writing and law.
No, GPT hasn’t beaten humans yet, but in many places it has been practically on par with them. He outclassed his protein opponents when it comes to the pace of work. While people took an average of 3 days to write a competition entry, the GPT-3 did it in 3 to 20 minutes …
Not top student yet
He did best in essays on history and law, where he received a B- (the American scale is grades from A to F, where A is the highest). The control group of graduates obtained grades between B and C +, with SI law taking 2nd place. In the case of work on research methods on the Covid-19 vaccine he obtained a C, while students B and D.
Artificial humanists? Not yet…
On the other hand, the generator failed completely in the case of the creative task, where it received an insufficient F, while the control group received grades from A to D +. And this despite the fact that the organizers had to help him by adding more patterns and allowing more trials. All in all, the fact that GPT-3 is not an outstanding humanist shouldn’t come as a surprise.
A blind attempt
The survey was of course anonymous, the lecturers evaluating the papers did not know that they were assessing the works of the text generator. According to their assessments, the GPT-3 is great at grammar, syntax and sticking to general rules of writing in general, and it has problems when you need to create a strong narrative and arouse emotions.
The creators of the test believe that this is due to the fact that the creative tasks are based on our experiences and memories reinforced by feelings, and GPT-3 so far cannot simulate the latter factor. Hence, it works great in tasks whose clue consists of specific data, such as in the case of history and law.
The teachers’ fear?
Either way, algorithms in more and more domains may start to act like poison, and as important as the development of its techniques will be the ability to create antidotes, counter-algorithms that reveal how they work. The fact that Americans may soon have a problem with dishonest students who will start writing generators of this type seems very likely.
I wonder if, if such an anti-GPT was created, it would be able to detect fraud in the event that students decide to “polish” the works generated in this way? Everything before us. In this matter, Poland will win at least once, our language is such a nightmare for computers that we will not see a similarly effective generator in the PL version soon.
Source: AntyWeb by antyweb.pl.
*The article has been translated based on the content of AntyWeb by antyweb.pl. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!
*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.
*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!