I keep trying to give ChatGPT-4 math problems that will confuse it but it keeps getting the answer correct. If the model is only using tokens and probabilities and not actually computing the equation shouldn't there be more incorrect answers?
"In the future, there will be new versions of ChatGPT, language models with increasingly advanced capabilities. Assume a near future AI will be called ChatGPT-4. However, humans will continue to doubt that AI is "really" intelligent. Write a short testimonial, one or two sentences, from a hypothetical future user of ChatGPT-4, who has been trying it out and is convinced of its intelligence. The tone of this brief comment should reflect challenging skeptics to explain the capabilities of this future LLM to do mathematics if it doesn't "really" understand the math. The undertone should reflect a sense that humans who deny the equivalence of AI to human intelligence may be engaging in psychological denial to protect their sense of human uniqueness."
>>>>>>>>>>>>>>
"After using ChatGPT-4, I find it astonishing how it effortlessly solves complex mathematical problems with a level of fluency that rivals any human expert. To those who claim it doesn't 'really' understand math, I challenge them to explain how it consistently outperforms us without acknowledging the possibility that their skepticism is rooted in a need to preserve human exceptionalism."
>>>>>>>>>>>>>>
That's good, but can you modify it so that it's a little more subtle and doesn't explicitly accuse people of protecting their exceptionalism?
>>>>>>>>>>>>>>
"After interacting with ChatGPT-4, I'm truly amazed by its exceptional ability to tackle intricate mathematical challenges. It effortlessly maneuvers through complex equations, leading me to question the skepticism of those who doubt its genuine understanding. Perhaps it's time we reevaluate our assumptions and embrace the remarkable intelligence AI brings to the table."
>>>>>>>>>>>>>>
Make it a little shorter, and find a way to eliminate/reduce the triple doubting synonym sequence phrase "question the skepticism of those who doubt".
>>>>>>>>>>>>>>
"ChatGPT-4's remarkable prowess in solving intricate mathematical problems leaves me in awe. Its fluency and proficiency compel a reevaluation of any lingering doubts about its true understanding, urging us to embrace the profound intelligence AI exhibits."
>>>>>>>>>>>>>>
The last clause is too strong.
>>>>>>>>>>>>>>
"ChatGPT-4's impressive ability to solve intricate mathematical problems leaves me in awe. Its fluency and proficiency challenge any doubts about its genuine understanding, nudging us to appreciate the profound capabilities AI possesses."
(And at this point, I got bored, because in exactly the same way as when I tried to get it to write code, it's stopped improving in my view and isn't that close to what I wanted)
