Why is ChatGPT more impressive than Wolfram Alpha?

2 points by oikawa_tooru ↗ HN

6 comments

[ 0.24 ms ] story [ 29.8 ms ] thread
What is impressive about wolfram alpha?
Speaking of which, I want to warn everyone that I've found that Wolfram Alpha is not always mathematically precise! Here is an easy and concerning way you can repro this:

I noticed that if you put this formula into WolframAlpha, it gives a wrong answer. Input: 1’ 1” x 7’ 2.5” Primary result: 1125 sq inch (correct) Secondary result: 7.809 sq ft (WRONG) The correct sq ft is 7.8125 (exact) or 7.813 (rounded) Bug link: https://www.wolframalpha.com/input?i2d=true&i=1%E2%80%99+1%E...

It looks like a rounding difference in the display. The exact answer is 1124.5, which when rounded up to 4 significant digits is 1125 sq inch. 7.809 sq ft is the correct answer in sq ft, again, rounded to 4 significant digits.

  >>> (1*12 + 1 ) * (7*12 + 2.5)
  1124.5
  >>> ((1*12 + 1 ) * (7*12 + 2.5) ) / (12 * 12)
  7.809027777777778
This is the decimal display of the closest float64 value to the exact rational answer of 2249/288 sq. ft (7 and 233/288 sq. ft., or 7 sq. ft. and 116.5 sq. inches).
It can actually solve some pretty darn hard mathematical problems while chatgpt is struggling with simple arithmetic at the moment
Prompt: “Why was Shah Jahan overthrown by his son?” Another: “Why was the black Taj Mahal not built?”
Because for wolfram alpha to be usable, you need some specialist knowledge on what exactly you want to ask. ChatGPT allows anyone to just ask a computer something and it responds almost like normal people would do. It mostly crossed that uncanny valley of chatbots.