Forking Mad+

OpenAI GPT fail

I was having a discussion with someone about AI earlier and ChatGPT. He claims it can solve simple problems 100% of the time. So I set it a basic question, while we sat chatting in the local bar!

I asked:

Mary has two sisters. Her sister, Joy, has two brothers. How many brothers does Mary have?

And the response:

Mary has one brother.

Here's the reasoning:

Thus, Mary has one brother

The reasoning is all correct, but the final conclusion does not follow the reasoning.

If it's not clear: Mary actually has 2 brothers; the same as her sister Joy!

The results were from GPT-4o. If I use GPT-41 if gets it correct.


Comments?

#AI