LLMs are trained on vast portions of the internet.
In that sense, they are essentially systems that mimic online content.
Therefore, what an AI generates often reflects the perspectives of the people who originally created the training data, rather than the true thoughts of the person prompting it.
I think there’s a small misunderstanding, though: while systems like GPS do account for relativistic effects at the clock level (and even with extremely precise atomic clocks and practical synchronization via NTP), this doesn’t mean we have a universal or perfectly shared notion of “the same time” across distributed nodes — especially once you consider network delays, clock drift, and faulty or unreachable nodes
In physics there is no absolute global time for spatially separated events, and in distributed systems this shows up as unavoidable uncertainty in synchronization.
Also, the FLP result isn’t about relativity or physical clocks at all — it’s a theoretical result about the impossibility of guaranteed consensus in fully asynchronous systems with failures.
So even with very accurate clocks and practical time bounds, distributed algorithms still have to explicitly deal with uncertainty and partial synchrony rather than assuming perfect global time.
A complete reference implementation showing how to build resilient microservices with guaranteed message delivery. Features MassTransit state machines, transactional outbox pattern, and automatic failure compensation across 4 services (Order, Payment, Inventory, Notification). Includes detailed docs on handling distributed transactions without 2PC.
In that sense, they are essentially systems that mimic online content.
Therefore, what an AI generates often reflects the perspectives of the people who originally created the training data, rather than the true thoughts of the person prompting it.