The real difference between float32 and float64

Question

I want to understand the actual difference between float16and float32in terms of the accuracy of the result. For example, Numpyallows you to select the range of the desired data type (np.float16, np.float32, np.float64). It bothers me that if I decide to go with float 16 to reserve memory and avoid possible overflow, will this lead to a loss of final results compared to float32, for example?

thank