Previous comment: GPT-4 failing with a math question:
https://chat.openai.com/share/53116ac3-0c44-410a-b367-c5df0a...
(I didn't discover this. I was just trying to confirm something that was reported elsewhere.)
Previous comment: GPT-4 failing with a math question:
https://chat.openai.com/share/53116ac3-0c44-410a-b367-c5df0a...
(I didn't discover this. I was just trying to confirm something that was reported elsewhere.)
not going to perform as well as GPT-4
Therefore, the largest fraction less than 1/2 with a numerator and denominator that are positive integers less than or equal to 10000 is 9999/10000
https://www.wolfram.com/wolfram-plugin-chatgpt/
Part of effectively using a tool like ChatGPT is understanding the boundaries and how to mitigate them. ChatGPT has lots of problems, but can still be incredibly useful.
Seriously though, if it can really reason I wonder if you feed it Peano’s Axioms will it from first principles be able to do arithmetic? Then again I guess it already know about Peano’s Axioms since it’s on Wikipedia, so the answer is “no”?
1. We're already finding uses for LLMs that go beyond what they've been trained to do. They act as if they have some model of the world. What is special about math?
2. It's impossible to understand the boundaries without probing them.
3. Here's ChatGPT's response after it called Wolfram Alpha: https://chat.openai.com/share/421fb625-ef57-40e5-bdee-b423b4...