Is GPT designed to cater to humans rather than give correct answers?

I used both Windows and the API to grade the same math problem involving solving an equation. Due to an OCR extraction error in the question stem, GPT provides answers that are inconsistent with the students’ answers. However, after correction, both the API and Windows show that the student’s answer aligns with the GPT-generated correct answer.

This is ridiculous.

thanks for the reply.It’s really a question of balance. However, in solving math problems at the elementary level, it should be precise. And prompt is indeed the key element.