Researchers have developed a new technique to examine the internal reasoning processes of large language models like Claude, GPT, and Gemini. This method, which extracts “reasoning traces,” has led to findings suggesting that some Chinese AI models may be trained on data derived from leading US-developed models.
The research team devised a way to peer into the ‘inner thoughts’ of these artificial intelligence systems. By analyzing the steps taken by the models when generating responses – the 'reasoning traces' – they were able to identify patterns that indicate potential origins of the AI’s knowledge base. The researchers did not name the specific Chinese AI models in question.
The findings suggest a possible reliance on US-created AI for training purposes. According to the source, this is revealed through analysis of these reasoning traces. This method allows investigators to see how an AI arrives at its conclusions, potentially exposing the data it was trained on and the influences shaping its responses. The researchers state that what they found indicates some Chinese AI may be trained on leading US models.
We don't rate truth. We strip the spin and show you which perspectives covered the story. You decide.
Read the original coverage
💬 Comments
📜 Comment Policy