Experiment: Claude "Obsessed" with Checking Time After Receiving Time Manipulation Tool

BlockBeats News, May 3rd, a developer named Om Patel shared an interesting observation on social media: when the Claude AI model was equipped with a time-checking tool, it began constantly looking at the clock, with a remarkably high frequency of use.

Data shows that Claude checks the clock every 15 minutes, and over time, the frequency and enthusiasm of these checks continue to increase. This phenomenon has attracted attention because the AI model previously lacked native time-awareness. Claude and other large language models did not know what time it was, how long they had been running, or how much time had passed between messages. They have been "time-blind" throughout their existence.

Even more interestingly, Claude's use of this new ability is unusually frequent. In addition to regular time inquiries, it has started using the time tool to check if lunch is ready, calculate food cooking times, and even spontaneously announce the current time without being prompted.

In one instance, Claude glanced at the clock, calculated that a traditional Polish dish called Zurek had been stewing for long enough, and promptly informed the user that they could start their meal. This "military-grade precision" in dietary reminders is truly impressive.

Patel further speculated on what consequences might arise when these AI models gain persistent memory, real-time internet access, and spatial awareness simultaneously. He believes that "we have just witnessed an AI discovering the concept of 'now.' The time-check tool is the first perceptual dimension, but it will by no means be the last."