Anthropic: Claude has developed the ability to 'think' internally
Anthropic has discovered that its Claude models have formed 'J-space,' a set of neural patterns enabling silent thought, similar to how human brains function. The company stated that J-space was not intentionally programmed or built by them, but rather appeared spontaneously throughout the training process. This allows Claude to perform internal reasoning without needing to output text, according to Anthropic.