“The Llama 2 paper states that reinforcement learning proved highly effective for language models, particularly given its cost and time effectiveness.”