International Journal of Mathematical Education in Science and Technology. They posed a 2,400-year-old mathematical conundrum to ChatGPT. The challenge involves enhancing the area of a square. In the initial account, the learner increased the length of every side of the square. Following a back-and-forth discussion, Socrates steers them towards employing the length of the original square’s diagonal as a basis for calculating each side instead. However, the goal isn’t merely to apply the mathematics accurately. Socrates aimed to reveal that the student already possessed the essential knowledge to compute the genuine answer through logical thinking. Scholars have debated this for centuries — questioning whether mathematical knowledge is inherent or acquired via reasoning and experience. But how do substantial language models (LLMs) such as ChatGPT address this?
Dr. Nadav Marco, a co-leader of the study from the Hebrew University of Jerusalem, collaborated with Andreas Stylianides, a Mathematics Education professor at Cambridge. They concluded that since ChatGPT is educated on textual data rather than images, if it can determine the correct answer, it bolsters the theory that mathematical capabilities and reasoning are cultivated rather than instinctual. Is this knowledge stored away to be “retrieved” or is it “created” through personal experiences? They assumed there was a slim possibility the chatbot would arrive at the correct conclusions. What transpired was that the bot adapted to discover a solution, even making a comparable error to the human learner — the bot mistakenly asserted that the diagonal could not be utilized and that there was no geometric solution.
<div class="slide-key image-holder gallery-image-holder credit-image-wrap " data-post-url="https://www.bgr.com/2002684/how-2400-year-old-greek-problem-chatgpt-ai-human-intelligence/" data-post-title="How A 2,400 Year Old Problem Shows Us How Close ChatGPT's AI Is To Human Intelligence" data-slide-num="1"
