Reflection is a prompting strategy used to improve AI systems' quality and success rate by having them reflect on their past actions and incorporate external information. This approach can help LLMs break out of purely System 1 "thinking" patterns and exhibit more System 2-like behavior. Reflection takes time, but it's worthwhile for knowledge-intensive tasks where response quality is more important than speed. Three examples are outlined: Basic Reflection, Reflexion by Shinn et al., and Language Agent Tree Search (LATS) by Zhou et al. LATS combines reflection/evaluation and search to achieve better overall task performance compared to similar techniques. It adopts a standard reinforcement learning framework, replacing RL agents, value functions, and optimizers with calls to an LLM. This approach can effectively use explicit reflections and web-based citations to improve response quality and adapt to complex tasks.