Don’t even know how to do junior high school questions, ChatGPT, Wenxin Yiyan, Claude showing chicken feet?

“The results I calculated with several models are different…”

On July 14, a group of friends used AI to help calculate a math problem: what is the volume in milliliters of a round platform with a height of 11cm, a diameter of 7.8cm at the upper bottom, and a diameter of 6.2cm at the bottom?

This netizen used Claude-2, GPT-4, and ChatGPT, and the results were: 3634.57 ml, 359.4 ml, and 469.3 ml.

Another group friend used Wenxin’s words to get a result of 64474.666666666635 milliliters.

“I can’t do the junior high school questions”, “Good guy, they are all different,” the group of friends commented in a hurry.

I also tested it with ChatGPT out of curiosity, and the result was 1436.08 ml.

The calculation steps given by ChatGPT are completely correct, the volume formula of the circular table = πh*(R^2+r^2+R*r)/3.

However, the calculated result is wrong.

I asked ChatGPT to answer again, and the result was 513.47 ml.

It’s outrageous, the calculation steps are completely correct, and the final result is different every time.

I also used the “AI” that comes with Baidu browser, which is a large model supported by Wenxin Yiyan.

The first result is: 193522.10746113118 ml

This is so wrong, I asked again and got the result: 1168.75 ml

Still not right, I asked again and the result was: 1099620ml

After repeated questioning, Baidu AI is no longer installed, and it is completely broken.

It was previously reported that GPT-4 scored full marks in MIT’s mathematics undergraduate degree exam, and it was later revealed that a large part of the test data set was contaminated. In other words, the model is like a student who was told the answer before the exam, which is blatant “cheating”.

It was also reported earlier that ChatGPT overturned while taking the mathematics test of the Chinese college entrance examination.

Large-scale models are undoubtedly a technology that has been sought after recently. However, the frequent cases of car rollovers seem to be what Zhang Tianrong, a former physicist and popular science writer, said. The essence of language models is the victory of probability theory. "The machine **, the converter makes a reasonable continuation of the input, and it is not difficult to understand the serious nonsense jokes.

If the large model is the victory of probability theory, then the awakening of artificial intelligence is far from coming.