<ul><li>AI models like ChatGPT historically play the word guessing game for generating text responses.</li><li>New generation models like DeepSeek’s R1 and OpenAI’s o3-mini are focusing on thinking and reasoning like humans.</li><li>A research paper in 2022 introduced the STaR method to teach models to develop rationales for their answers, encouraging thinking over word generation.</li><li>STaR method rewards models for providing the right answer with a good rationale, aiming to refine their reasoning abilities.</li></ul>

How do AI models think and reason?

Discover more