Good day,
I have seen a few examples of how to prompt for complex math word problems using GPT on the openai cookbook GitHub, which is a repository of useful tips and tricks for working with GPT models. You can find it here: GitHub - openai/openai-cookbook: Examples and guides for using the OpenAI API. One thing I learned from there is that you need to be very specific and clear when asking GPT to solve math problems, and use proper symbols and units. For example, if you want to calculate the force of gravity between two objects, you could prompt GPT like this:
Calculate the force of gravity between two objects with masses m1 = 10 kg and m2 = 5 kg, separated by a distance of r = 2 m. Use the universal gravitational constant G = 6.67 x 10^-11 N m^2 / kg^2.
One interesting thing to explore that I started playing with is instead of asking these complex questions, you can get GPT to create simulations for your data sets, and then use the specific model to generate the answers. For example, if you have a data set of measurements of the position and velocity of a projectile over time, you can get GPT to fit a quadratic function to the data and then use that function to predict the maximum height, range, and time of flight of the projectile. I have been using the GPT-3 model for this purpose, but you could also try other models like GPT-J or GPT-Neo. This way, you can avoid the problem of unit conversion and loss of decimal places that sometimes occurs in physics based mathematical problems.