That was a good question. It wasn’t asking ChatGPT to regurgitate knowledge. It was asking it to solve a simple problem. That’s not what ChatGPT was trained for.
ChatGPT is like an arts student. It may know lots about engineering, but ask it to solve a simple engineering problem and it will fall on its face.
It's because it's a probabilistic text generator. This is a dead-end for AI because it can't attempt to determine what is true or accurate, simply the likely format for the output. It has no consideration for what is true/false or logical/illogical. e.g. if you asked it to describe a legal dispute between Isaac Newton and Stephen Hawking, it would blithely invent a court case of Newton v Hawking rather than state "They're both dead, they can't sue each other".
Or in the case of pigs, would (badly) attempt to count feet in the same manner of the lowest common denominator of the internet. Since the lowest common denominator of the internet are retarded normies, it'll count just as badly as they would. Even if a calculator or image recognition ML algorithm would easily count them.
The ML model used by wolfram alpha tends to fill this gap of numeric accuracy, though not all the time. I tried entering the same question into that but it can't interpret the wording well enough to start the math.
That company is current working with OpenAI to form a bridge between those two systems. After that it would be interesting to see how accurate it responds.
(post is archived)