Today, a neural network based on NLP (natural language processing algorithms) called the GPT-3 transform network is the most advanced. It allows healthcare providers to answer people’s most important questions. The neural network of the previous generation called GPT-2 has a hundred times fewer parameters.
But even those transformers, which are the most advanced, cannot understand the meaning of the generated text. A huge amount of data and computational power are needed for the procedure of their «training». Another difficulty is that not all datasets are suitable for training neural networks, since many sources from the Internet may contain false information or manipulations.
Now developers are trying to achieve the successful creation of such multimodal systems that can work with touch with text recognition. This will enable professionals to process data and find solutions more efficiently.