Recent advancements in artificial intelligence have led to the development of 'Thinking LLMs' (Large Language Models) that exhibit human-like reasoning and thought processes. Research from Google DeepMind and Stanford has introduced groundbreaking methods for training these models to perform a wide range of tasks beyond traditional reasoning, including marketing, health, and general knowledge. A notable innovation is the Thought Preference Optimization (TPO), which trains LLMs to think and respond for all instruction-following tasks, surpassing the performance of models like GPT-4 and Llama3-70b on benchmarks such as AlpacaEval and ArenaHard. Meta presents these findings, highlighting gains from thinking on even non-reasoning categories. Additionally, Rational Metareasoning for LLMs has been proposed to enable these models to use reasoning only when necessary, potentially reducing operational costs. Wu et al. have contributed to this research, emphasizing the potential for LLMs to revolutionize various fields by optimizing their internal thought processes through reinforcement learning.