While learning about LLM, I came to understand that generalization from the training data and learning to find context patterns are among the factors that make an LLM sound more human. So, transformer architecture (a black box, as I do not know its internal workings) is used for this purpose of generalization, context matching, etc. For example, if-then pattern, cause-effect pattern, coherence resolution, and so on. My doubt is, can LSTM model...
Originally asked by shree on Stack Exchange (ai) (CC BY-SA 4.0).
