This TED Talk features an interview with OpenAI research scientist Noam Brown, who discusses his research on AI's ability to "think." Brown explains how incorporating "System 2 thinking" (slower, more deliberate thought processes) into AI models significantly improves performance, comparable to massive increases in model size and training data. He uses his experience developing poker-playing AIs as a case study, highlighting how allowing the AI to deliberate for 20 seconds yielded performance gains equivalent to a 100,000x increase in model scale and training time. This suggests a new paradigm for AI development, focusing on enhancing "thinking time" rather than solely scaling computational power. The talk concludes with the introduction of OpenAI's new language model, o1, which incorporates this "System 2 thinking," demonstrating the practical application of this research.
