“Scary”: ChatGPT passed the Bavarian Abitur at the second attempt

experiment
ChatGPT passed the Bavarian Abitur at the second attempt: That’s why the AI ​​is now “smarter”

The artificial intelligence ChatGPT tries the Bavarian Abitur tasks 2023 (symbol image)

© Michael Gstettenbauer / Imago Images

ChatGPT has blossomed from problem student to nerd: within a few months, the chatbot has improved significantly in solving high school exams. But why is that?

German, history, math, ethics and computer science – the artificial intelligence (AI) ChatGPT had to prove itself in these five subjects and solve high school exam tasks. As recently as February, the chatbot had failed in last year’s exam tasks, which the computer linguists of Bayerischer Rundfunk (BR) had presented to it. The program received insufficient marks and failed in almost every subject.

In another attempt, the AIs were recently confronted with this year’s exam questions. It started with the tasks of the history high school diploma in 2023. Here, among other things, the change in society from the 15th to the 19th century with the importance of the railway and the position of the Jewish population were queried. Judith Bruniecki, history teacher at the high school in Munich-Moosach, gave ChatGPT a grade of two for the solutions provided by ChatGPT. “The AI ​​has difficulties especially when it comes to assessing something, weighing it up and maybe even thinking outside the box,” Bruniecki told BR. The program received a three plus for solving the history exams in 2022.

ChatGPT works more thoroughly and “at a consistently very high level”

The AI ​​also improved in ethics – from a four minus to a two minus. “Compared to the last time, when the AI ​​only just passed, this is a big step forward,” says Winfried Kober, ethics teacher at the Ludwig-Thoma-Gymnasium in Prien. Overall, the answers would be well-formulated and clearly geared to the task at hand. Although Kober found that there was a lack of technical terms in some places, he nevertheless recognized an overall increase in performance.

The chatbot also showed clear improvements in German. In February, a German teacher commented on his performance with the words “lots of drivel, little substance” and gave him three points. The AI ​​would have failed the exam. When asked to interpret a parable by Ilse Aichinger, ChatGPT did a good job. “I think the language is of a consistently high standard,” says Christoph Willing, German teacher at the St. Anna High School in Munich. He awarded the AI ​​ten points for its thorough and in-depth text interpretation, which corresponds to a two minus.

Teacher finds AI performance improvement ‘bit scary’

The chatbot was just as convincing in math. “While in the last round you sometimes read incomprehensible gibberish, this time the answers are very well structured and the solution is almost always efficient and easy to understand,” says Thomas Spindler, mathematics teacher at the Luisen- High school in Munich.

However, the AI ​​made the most serious increase in the subject of computer science. She failed the Abitur assignments from 2022 and failed with two points. Now she got a total of eleven points, a grade of two. Hermann Kees, computer science teacher at the Joseph-Bernhard-Gymnasium in Türkheim thinks: “The AI ​​has done a lot right and I find this development almost a bit scary.”


Experiment: ChatGPT passed the Bavarian Abitur at the second attempt: That's why the AI ​​is now "smarter"

Progress through new model

But why is it that the AI ​​is doing better across the board in this year’s high school exams? In February, the model based on GPT 3.5 still solved the problems. The now available GPT 4.0 can process comparatively more data and is more accurate in the text output. According to BR, GPT 4.0 is “more powerful and, if you will, smarter”.

Sources: BRMirror

source site-5