A test carried out by DeltaFolha showed that ChatGPTartificial intelligence developed OpenAI, will perform better than 80% of students on Enem, the National Secondary Education Examination.. However, in mathematics, the robot did not excel, beating only 27% of those who passed the exam.

The analysis was based on AI responses to tests conducted over the past five years. ChatGPT answered 1290 questions, which is rare in Portuguese at this scale. Two exam applications each year were considered, which contain completely different questions.

The robot will have an average score of 612.3 on Enem’s objective tests, surpassing 98.9% of students in the liberal arts and 95.3% in languages ​​and codes. Overall, he would have outperformed 78.9% of applicants. However, his academic performance in the exact sciences would have become a serious obstacle to enrolling in controversial courses at the country’s main federal universities.

Analysis

The analysis took into account the calculation that each course takes (the weight of disciplines varies depending on the degree). According to Sisu (Uniform Selection System), this score guarantees admission to 63 of the 938 options listed by the top ten federal universities in the 2019 Fogli University Rankings.

Math test scores were the lowest, with the robot averaging 443.1 points compared to 527.1 for real candidates. He got between 13.6% and 27.3% of questions right on every app. In contrast, the liberal arts were AI’s strengths, with a five-year average of 725.3, higher than the students who scored 523.3.

As for the essay, in the report, the system asked to write an essay in the same wording as the test. The average score of the robot was 700, which is better than the 68% of students who had an average of 613 points. Adding the written score to the average objective test score in 2021, Enem’s ChatGPT score was 608.7, a 79% improvement. students this year.

In short, the DeltaFolha analysis showed that ChatGPT would do well at Enem, but that its poor math performance would be a barrier to entry into controversial courses at the country’s major federal universities.

Methodology

Using the Response Task Theory, a mathematical model adopted by Enem that predicts items calibrated according to discrimination, difficulty, and probability of random correctness parameters, Folja calculated the final ChatGPT score. To replicate Inep’s methodology, the robot answered each question only once, without previous examples, indicating the option it thought was correct. Reader versions of Enem were used, since the technology does not interpret images, and GPT was set to limit possible “clutter”.

Mundo Conectado Deal Center: Selection of Discounts and Lowest Prices
Best deals on electronics, cell phones, TVs, soundbars, drones and more

Source: Folha de São Paulo.

…..