Absolutely not. Here's why:
AI (at least LLMs in IM) is a fad. Much like how everyone thought "blockchains are the futureeeeee". What replaced blockchains? They served some purpose, right? Simple databases. There's literally 0 reason to spend gigawatts of power for something to serve the same purpose as a database, which is pretty energy-lite.
But what about ChatGPT?
Right now, ChatGPT is like, actually at its worst quality ever, especially since they updated their dataset to something more recent. Either that or I suck at prompt engineering. But if I sucked at prompt engineering, why does switching to OpenAI API GPT-3.5-turbo-instruct solve all of my problems? I can think of a few reasons:
1. I suck at ChitChat. Turbo instruct is to take instructions, Chat is for conversation. No idea why OpenAI thinks people want Chat over Instruct. When I instruct a machine to do something, "I'm sorry, but..." is unacceptable. In Chat, that seems to be the most common response these days.
2. OpenAI is nerfing it.
3. LLMs will become outdated and inaccurate. Think about it. Since the release of ChatGPT, how much of the internet started using AI content? For every 1 human written article in 2023, there are at least 400 AI written ones posted online. AI content is really bad at training itself. So making larger LLMs is nearly impossible without mixing in a ton of AI content. I've tried other LLMs and the newer ones have the same problems as ChatGPT or GPT-4.
4. It's energy intensive. The environmental impact of LLMs is incredibly high compared to, say, an article spinner with grammar checker. I think a good article spinner makes better content than ChatGPT, at least, it did before ChatGPT was the primary source of content.
5. LLMs are bad at math. Machine Learning is really good at math, and the biggest competitor to calculator web apps is Wolfram Alpha and Google. Machine Learning also isn't terribly energy intensive, either. However, ML in math applications is not an LLM.
So to answer the question, no, ChatGPT/any LLM will not replace calculator apps. It's too energy intensive to run 1 + 1 through an LLM just for it to say the answer is 7 or something stupid. However, Google's rich results (which can include a calculator) and HUGE sites like Wolfram Alpha will make it extremely difficult to compete in the calculator niche. If you've got a feature that these calculators don't have, lean into that heavily.