Generally speaking, it's true. AI is quite bad at arithmetic. But as AI started learning to use tools things can be different. Haven't really looked into this but I think ChatGPT with WolframAlpha plugin (https://www.wolfram.com/wolfram-plugin-chatgpt/) can be helpful.
People won't use generative models when they already have a clear plan about what they're about to calculate - they'll use a calculator. When employing generative models, we're more like asking the AI for some insights on "how to compute" or "what to compute". Automatically performing these calculations and obtaining the result is more like a bonus.
> Mathematica and wolfram are in a different pricing tier, so hissab is like a poor man's mathematica.
Agreed. Looking forward to exciting features!