AI Model GPT-3 Demonstrates Remarkable Reasoning on Par with College Undergraduate Students


With Astonishing Results, Researchers Confirm GPT-3’s Comparable Performance to Humans, Including Similar Mistakes Made

According to a study conducted by researchers from the University of California – Los Angeles (UCLA), the popular AI-powered tool GPT-3 demonstrated reasoning abilities comparable to college undergraduate students. The researchers asked GPT-3 to solve reasoning problems typical of intelligence and standardized tests, including SAT analogy questions.

In the shape prediction test, GPT-3 correctly solved 80 percent of the problems, outperforming the human test-takers’ average score of just below 60 percent. Surprisingly, the AI also made similar mistakes as the human participants.

When tackling SAT analogy questions, GPT-3 performed better than the average human score. Analogical reasoning involves solving unfamiliar problems by comparing them to familiar ones and extending those solutions to new scenarios.

However, the AI showed less proficiency in solving analogies based on short stories. In these cases, the problems required reading one passage and identifying another story conveying the same meaning, an area where human students performed better.

The researchers were amazed at GPT-3’s reasoning capabilities, considering that language learning models are primarily focused on word prediction. While the AI demonstrated promising reasoning abilities, the researchers did not have access to GPT-3’s inner workings, guarded by its creator, OpenAI. Thus, they are unsure if GPT-3’s reasoning mimics human thought or if it represents an entirely new form of artificial intelligence.

As the researchers continue their exploration, they are curious to determine if GPT-3 genuinely reasons like humans or if it employs a different, novel approach. This study opens new avenues for understanding the potential capabilities and advancements of AI technology, and it may pave the way for future research in this groundbreaking field. However, it is important to note that the claim of 100% accuracy in GPT-3’s reasoning abilities was not mentioned in the study and remains unverified.

Leave a Reply

Your email address will not be published. Required fields are marked *