Can AI think outside the box? Researchers recently put large language models (LLMs) to the test with brain teasers designed to challenge common sense and require lateral thinking. These puzzles, divided into sentence-based and word-based challenges, aimed to see if AI could move beyond logical, linear reasoning. The study used a powerful LLM called Gemini 1.0 Pro and tested various prompting methods, including zero-shot, few-shot, and dynamic few-shot learning. Researchers also experimented with providing the model with explanations and reasoning behind correct answers. The results were intriguing. While the LLM performed significantly better than baseline models, it still lagged behind human performance. Interestingly, providing the model with the definition of a brain teaser improved its ability to reconstruct the context of the puzzle. Adding more examples in few-shot learning also boosted performance, highlighting the importance of context for AI. However, adding self-generated reasoning didn't always help, sometimes creating a trade-off between different types of questions. For word puzzles, static examples and reasoning proved surprisingly effective. This research shows that while LLMs can tackle lateral thinking puzzles, there's still a gap between AI and human ingenuity. Future research could explore how to better equip LLMs with the kind of flexible, creative thinking needed to truly crack these challenging brain teasers. This could involve incorporating more diverse training data, developing new prompting strategies, or even integrating cognitive architectures inspired by the human brain.