@BasedLunatic
Math will never work with current LLM AIs, because math is the language of logic and reason, and LLMs can't do precise reasoning like that. By the time you get these AIs to do math correctly, they'd barely be any better than humans.

Computers do math well when you program them to use their precise binary processors to do precision things, aka raw computational math. Send 0010 + 0011 to the CPU and get 0101 in return.
LLM AIs are not that. AIs are an attempt to get variety, originality, aka imprecision from binary precise processors. So you take a precision instrument, force it into an imprecise process, and then you expect it again to get a precise math result. You're just using all of it wrong.

Ideally, you'd want to make the AI be able to identify when a user is asking for a math problem, and make it manually send the parameters to a Calculator app, and just plop the result in chat. I doubt right now that any AI is doing this. They're just sending everything in the LLM and hoping magic correct math comes out.