Recent research reveals that training chatbot models like GPT-3 consumes a significant amount of water, equivalent to filling a nuclear reactor's cooling tower. Specifically, GPT-3 alone uses approximately 185,000 gallons or 700,000 liters of water, meaning that every conversation with chat GPT results in the wastage of a large bottle of fresh water. The study highlights environmental concerns and emphasizes the impact on drought-prone regions like Texas, where tech companies have their servers and cloud facilities. Although the study focuses on evaporated or lost water rather than recycled water, the speaker suggests that it may be a lesser concern compared to other issues like nuclear waste. Nevertheless, as more AI models are developed, monitoring and scaling water usage responsibly becomes crucial alongside considering the costs of employees, training, and data.