The key to fixing that robotic tone is to train your LLMs on conversational data from actual dialogues rather than just formal text. In our experience, we found that supplementing training datasets with casual chat logs or support interactions can significantly humanize responses. Another effective method is to adjust your prompt engineering to use informal language and context for more natural outputs. - Ali Muwwakkil (ali-muwwakkil on LinkedIn)