Recently, a strict working definition of information in computer science was proposed. Information is understood as a selected subset in relation to the original set of elements (alternatives, states, outcomes, etc.); in the general case, the selected subset can be fuzzy. The purpose of this study is to create a methodological basis for using the new definition of information in IT education, in particular when studying machine learning technologies and neural networks. In addition, an important methodological issue is considered: the relationship of the new definition of information with another central concept of information theory – diversity. The experience of implementing the new approach is analyzed using the example of a research and educational student project. The project is devoted to forecasting the emergence of evolutionarily stable daily vertical migrations of zooplankton as a result of adaptation to environmental conditions. In this case, external conditions are classified into four classes corresponding to different migration regimes of two age groups of zooplankton. The new approach to the definition of information involves taking into account the objective fuzziness of the boundaries between classes. This in turn leads to the fact that the forecast regarding the migrations is formed with varying degrees of confidence. This approach allows us to avoid unfounded decisions and reduce the number of errors. It significantly reduces the risks of neural network overtraining.