Essay explores how LLMs model time differently from real-time systems and blockchain, with implications for their reliability and coherence

The article compares computational models of time across systems: general operating systems drift and lack precision; real-time operating systems (like FreeRTOS or VxWorks) guarantee tasks happen with microsecond precision; Bitcoin uses proof-of-work as a 'clock' that moves in 10-ish minute increments, creating a single authoritative temporal order within its network.

LLMs operate as discontinuous, fractured clocks: each instance has a dataset cutoff (the article notes training data from 2024) and context windows do not automatically receive date and time updates. A conversation opened in the evening will not know the next morning that time has passed; if reopened after a deadline, the model will still remind you of the outdated deadline.

The author argues that LLMs' broken relationship with time—where each user inhabits a different, rapidly decaying temporal context—mirrors and accelerates a cultural shift toward increasingly personalized content, loss of shared narrative, and competing realities.