Large language models have been undeniably exciting with a myriad of potential product applications. In this talk, we'll describe the process for creating the improved LLaMA-2 family of LLMs and the techniques we used to convert these to high-quality, safe dialogue agents