Researchers have developed a technique to extract hidden “reasoning traces” from leading AI systems, including Claude, GPT, and Gemini, revealing detailed step-by-step internal processes as the models solve complex tasks. By analyzing these internal traces, the scientists compared problem-solving styles across different systems. They observed striking similarities between some Chinese-developed models and US models, suggesting possible reliance on outputs from established Western systems. The findings raise fresh questions about intellectual property, safety, and competitiveness in the global AI race. Policymakers and companies may push for stricter transparency standards, dataset disclosures, and technical safeguards against covert model-to-model training.
This update represents a notable development in the Ai sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.