Challenging AI to be Smarter: Stanford University Reveals New Tests for Improvement

Stanford University researchers have emphasized the need to redesign tests to challenge AI and help it become smarter and more advanced. These tests have shown that AI abilities are comparable to an ordinary person in various tasks, but there is still much room for improvement.

One area where AI has made significant progress is in image classification and text understanding in English. The HAI Research Institute at Stanford University conducted a review of different versions of AI and assessed their capabilities. The 500-page report highlighted that the best AI systems can compete with humans in certain areas or act as assistants in performing tasks.

In particular, Open AI’s AI system has shown remarkable progress in solving mathematical problems on the Stanford MATH test, while also demonstrating promising results in predicting and assessing real-world scenarios such as a cat jumping off a table.

Despite these advancements, challenges remain. For example, popular large language models like BERT are still prone to errors such as “hallucinations.” To reach its full potential, AI needs to further improve in areas like mathematics and design.

Stanford University is working on developing new tests that can accurately compare AI systems and identify areas where AI capabilities differ from human abilities. The introduction of new language models like GPT-5 is expected to shape future tests and push the boundaries of AI capabilities even further.

Researchers are optimistic about the potential for AI to excel in various tasks while acknowledging the importance of identifying and leveraging the unique strengths of human intelligence. As AI continues to evolve and improve, we can expect to see even more impressive achievements from these powerful machines.