Can you tell the difference between a human and a machine in a conversation? The Turing test, designed to assess a machine's ability to exhibit human-like intelligence, has long been a benchmark in AI. Traditional Turing tests involve a rigid back-and-forth dialogue structure and require constant human oversight. But what if we could make the test more dynamic and realistic, mimicking the natural flow of human conversation? Researchers have introduced the "Self-Directed Turing Test," a new framework that allows for more complex, multi-message exchanges, much like how we text and message each other daily. This approach also reduces the need for constant human involvement by letting the AI, or Large Language Model (LLM), "self-direct" parts of the test by generating simulated conversations with humans. After generating a long sequence of this pseudo-dialogue, the LLM engages in a shorter conversation with a real human. This interaction is then compared to a human-human conversation on the same topic, and judges try to spot the AI. The results? While LLMs like GPT-4 initially performed well in shorter conversations, their ability to convincingly mimic human responses declined as the conversations became longer. This new test design highlights the challenges LLMs face in maintaining consistent, human-like behavior over extended interactions. It also emphasizes the need for more robust evaluation methods that capture the nuances of human communication and push the boundaries of AI development.