A New Trick Reveals AI Models’ Inner Thoughts

Researchers developed a method to extract "reasoning traces" from Claude, GPT, and Gemini. Their findings suggest that some Chinese AI may be trained on leading US models. This method involves analyzing how models make decisions, providing insights into their learning and training mechanisms. Data from multiple models were used to identify similarities in patterns. These results highlight the importance of understanding the influence of American models on AI development in China. Such dynamics may impact global competition in the AI sector.