A new study found that ChatGPT AI, a large language model, can outperform undergraduates at solving Scholastic Aptitude Test (SAT) reasoning problems.
Research conducted by the University of California, Los Angles (UCLA) shows the problem-solving capabilities of the Open AI language model.
ChatGPT AI Outperforms Undergraduates at SAT Problem Solving
As per The Guardian, recent research found that the AI chatbot is an impressive academic contender, solving problems better than an average undergraduate student.
The study compared the performance of ChatGPT with that of real undergraduate students, and guess what? The results are pretty mind-blowing: Not only did the chatbot match these young minds' performances, but in some cases, it even surpassed their abilities. A study, published in the journal Nature Human Behaviour, found that ChatGPT correctly answered 80 percent of the SAT reasoning problems. On the other hand, the average undergraduate scored just below 60 percent.
The researchers believe that ChatGPT's ability to solve SAT reasoning problems is due to its ability to process information and draw inferences similar to how humans do. The chatbot also identified patterns and relationships between words and concepts, utilizing this information to solve the problems.
The study's findings have implications for the future of artificial intelligence. It suggests that AI models are becoming increasingly capable of solving complex problems that were once thought to be the exclusive domain of humans.
How Good is ChatGPT?
The study's lead author Taylor Webb stressed that ChatGPT still struggled with achieving human-level intelligence. Webb says the chatbot is lagging with mathematical reasoning, problem-solving, and even social interactions. Despite that, the language model is making significant progress these days.
The postdoctoral researcher in psychology, Webb, points out that ChatGPT is "definitely not full general human-level intelligence." Yet, the study's lead author acknowledges that the AI chatbot "has definitely made progress in a particular area."
According to FagenWasanni.com, the research shows that ChatGPT, using GPT-3, had trouble matching prose passages derived from various short stories, which essentially had the exact meaning.
But on the flip side, its successor, the more capable GPT-4, now only available to ChatGPT Plus subscribers, is better than its predecessor in matching these same-meaning passages.
Forbes reports that the Microsoft-backed OpenAI debuted ChatGPT last November 30 as a research preview. The chatbot quickly gained traction, outperforming the rise of social media giants like TikTok and Instagram. In less than a year, the revolutionary chatbot garnered roughly 100 million users globally.
Related Article : OpenAI Launches New Feature Allowing Users To Control ChatGPT Behavior