I found a weird thing in ChatGPT

https://chat.openai.com/chat/a0e92a0f-940a-442b-b6a7-d27e3e085459

by mrjay42 • 4 years ago

3 6 4 years ago

chat.openai.com

mrjay423• 4 years ago

If you ask chat GPT whether 2147462299 is prime, it'll output a wrong result.

And after discussing with the AI I got this response: "I apologize for any confusion my previous response may have caused. You are correct that 2147462299 is not divisible by 137, 73, or 41. In fact, 2147462299 is a prime number.

To determine if a number is prime, you can use a variety of methods, including trial division and the sieve of Eratosthenes. These methods involve dividing the number in question by a series of smaller numbers, typically starting with 2 and ending with the square root of the number being tested. If the number being tested is not divisible by any of these smaller numbers, it is likely to be prime.

It is also possible to use specialized software or online tools to determine if a number is prime. Some of these tools use more advanced methods, such as the AKS (Agrawal–Kayal–Saxena) primality test, which can determine with high accuracy whether a number is prime or composite.

I apologize for any mistake I made in my previous response. Please let me know if you have any further questions."

uberman• 4 years ago

How or why is this weird?

I think I have now reached my fill of posts showcasing what this tool said.

mrjay423• 4 years ago

It's weird because:

Recognizing a prime number is an easy task

Recognizing a prime number is an easy task for a computer

Providing erroneous factors is an easy thing to check

I am NOT "showcasing", to the contrary I am showing that this tool is weirdly inaccurate on typical things that any computer would normally compute easily.

I acknowledge that I used the word "weirdly" instead of "surprisingly".

dflock• 4 years ago

It gets stuff wrong all the time - it has no knowledge or understanding - it's just a very convincing stochastic parrot.

mrjay423• 4 years ago

Feels like it indeed

jaclaz• 4 years ago

But then it must have heard (or read) somewhere that that particular number is not prime, and it is just repeating that (wrong) statement without checking it.

So the real question is whether, after having verified that the statement is false and having apologized about the error, can it autocorrect its "knowledge" and not repeat the mistake?

I found a weird thing in ChatGPT

Related Stories