Spacerouter

I had a little thought experiment: we know that generative AI is not very good at math. So what if I could build some “middleware” for AI that sends math requests to Wolfram Alpha, but keeps generative AI for the things it is good at. I built a neural network classifier that can determine whether a piece of text is likely to be a math question or not. The answer is that it does seem to be better than ChatGPT without the middleware. However, Wolfram Alpha cannot answer every math question either. This can be an advantage or disadvantage depending on the way you see it. If it can’t be definitively solved, it won’t solve at all. However, there’s a chance you’re missing out on a solution that might or might not be correct. The biggest challenge was training the neural network. There is a lot of work to be done to increase the accuracy. In the future, it would be interesting to improve the neural network to classify more types of information and send them out to various third parties.