The upgraded AI-powered language model, dubbed GPT-4, scored 297 on the bar exam, outperforming the average score by 7%.
A professor and employees of the Casetext law firm conducted the experiment. According to the researchers, the points scored are enough to get admission to practice law in most US states.
The exam consisted of multiple-choice tests, essay questions, and quizzes on certain types of legal writing. The standard exam lasts two days, but the language model passed it in a shorter period.
The authors of the experiment believe that the test results depend on the version of AI. Thus, the more advanced the version is, the more points the AI can score. The researchers' previous experiment, in which another (earlier) ChatGPT language model failed to pass the bar exam, proves this. Artificial intelligence GPT-4, unlike its predecessor, correctly answered 76% of multiple-choice questions, while ChatGPT gave 50% of correct answers. Professor Daniel Martin Katz from Chicago-Kent College of Law, who tested the language model, said he was especially surprised by the AI’s ability to write engaging and relevant essays.
Share this with your friends!