Grading artificial intelligence

30.03.23 | Artificial intelligence company OpenAI’s newest model is doing remarkably well in numerous simulated exams. For instance, GPT-4 scored in the 90th percentile at exams in the fields of law, biology and microeconomics. It did less well, scoring at or below the 50th percentile, in English literature, calculus and programming. However, the improvement from the previous version GPT-3.5 is significant. So, although GPT-4 is less capable than humans in many real-world scenarios, it already does quite well on various professional and academic benchmarks. Intriguingly, the AI model also did well on a sommelier test, maybe because it wasn’t drinking during the exam and the humans were.

Source: OpenAI, 2023.