Possibilities ChatGPT impressed everyone, but the OpenAI AI chatbot is far from perfect. Its developers know this, and that is why they are working hard to ensure that its evolution continues to happen by leaps and bounds. However, there is one area that has proven to be particularly problematic for the platform: mathematics.
The situation is quite striking when one considers that one of the earliest and most primitive uses of computation was for mathematical purposes. And we are talking not only about arithmetic and geometry, but also about their implementation in even more specific industries.
The problem with ChatGPT with mathematics is not that it doesn’t work like a calculator. In fact, if you enter some simple operation in numeric format, it will probably work correctly. It is not infallible, but usually achieves the desired result. However, if you choose a natural language statement, the story changes drastically.
Recent report from Wall Street Magazine reveals this situation based on an article published by Professor Paul T. von Hippel of the University of Texas at Austin. The above has focused on ChatGPT’s problems when dealing with, for example, the Pythagorean theorem.
Although the chatbot was able to define it and provide examples, it is not without flaws. In fact, the expert noted that ChatGPT entered wrong digit when squaring a number. Something that should be very easy for a computer to calculate.
ChatGPT has serious math problems
Apart from what was explained by Prof. von Hippel when ChatGPT revealed content of a certain complexity, the chatbot still cannot solve some tasks when they are given in natural language. The above report Wall Street Magazine includes several examples of such a situation, but we also tested it ourselves.
“There are 6 bananas and 2 kg of apples in a box. If each apple weighs 200 grams, how many fruits are in the box? I asked the bot, and his answer puzzled:
Apples 2 kg, which is equal to 2 * 1000 = 2000 grams.
6 bananas means 6 bananas in total.
Total box 2000 + 600 = 2600 grams or 2.6 kg of fruit.
Of course what I was looking for is what ChatGPT answered me which is in general there were 16 fruits in the box: 6 bananas and 10 apples. The chatbot correctly calculated that 2 kg of apples equaled 2,000 grams, but never divided them by 200 grams of each apple to determine how many units were in the box. bye with bananas decided to assume (without explaining why) that they weighed 600 gramsat the rate of 100 grams each.
The answer that there were 2.6 kg of fruit in the box may not be completely wrong, assuming that the bananas actually weigh as much as the bot calculated. But it wasn’t what he expected. And ChatGPT invented a variable – the weight of bananas – to get the final result.
After trying different combinations of similar statements, always getting results in the same format, I decided to state what I wanted to know. how many pieces of fruit were in the box. It was there that ChatGPT solved the problem correctly.
Apples 2 kg, which is equal to 2 * 1000 / 200 = 10 apples.
6 bananas means 6 bananas in total.
There are 10 + 6 = 16 pieces of fruit in the box.
problems to solve
At the same time, a big problem with ChatGPT with mathematics is the impossibility of 100% understanding of statements if they are not formulated in the way they consider correct. A situation that makes the process less natural, especially if the platform is considered can serve as a teaching aid in the future.
Something that is in line with what was revealed by von Hippel: “Could ChatGPT provide comments and answer math questions in a more personalized and natural way? The answer, for now, is no. Although ChatGPT may talk about mathematics superficially, does not “understand” mathematics in any real depth. Cannot correct mathematical errors, often introduces his own errors. And sometimes he makes inexplicable math errors that wouldn’t show up in a normal spreadsheet or handheld calculator.”
Source: Hiper Textual
