A recent study has revealed that ChatGPT, an AI language model, shows problem-solving abilities on par with or even better than college students. The researchers tested GPT-3, the current version of ChatGPT, on reasoning problems similar to those found in intelligence tests and college entrance exams like the SAT. The model performed impressively well, outscoring the average score of human participants, who were 40 college students from UCLA.
In one test, GPT-3 had to predict the next image in a complex shape arrangement, and it solved 80 percent of the problems correctly, even when facing challenging SAT analogy questions. However, there were certain areas where GPT-3 struggled, such as understanding physical space, social interactions, and math reasoning.
The study’s lead author, Taylor Webb, clarified that while GPT-3 demonstrated strong pattern recognition and inference abilities, it did not reach human-level intelligence or artificial general intelligence. The researchers are intrigued by GPT-3’s unique reasoning process, which differs from how humans learn, and they are eager to uncover if it represents a new form of artificial intelligence.
In conclusion, the study showcases ChatGPT’s remarkable problem-solving capabilities while underlining the need for further research to understand the underlying mechanisms of its reasoning process.
