An economics professor at George Mason University gave ChatGPT questions from midterms in January, where the bot scored 31 out of 100 on the test. The result was far below the 50% median grade, and supported the professor’s belief that AI could not understand simple economics concepts.
The professor bet that an AI model would not perform well for years, so he was shocked when ChatGPT-4, released earlier this month, scored a 73% on the same test. He claims the bot gave clear answers and understood concepts it had previously struggled with. These improvements in GPT models could pose difficulties for classroom exams.
Read more at Business Insider.
Comments