The difference is not intelligence and it is not scale. It is that one system is asked a question and the other is never not being asked. A living predictor does not receive input. It has a standing prompt.
Navigation: ‹ Why Next-Token Prediction Is Prediction-Error Minimization (And Where It Isn't) · Who Owns the Loss Function ›
A language model can run in a loop.
This matters, because the simple picture is too simple. You type a prompt, the model produces a token, that token becomes part of the context, the model produces another token, and so on until the answer is complete. If the model is placed inside an agent harness, the loop can grow larger: read, write, call a tool, observe the result, plan again, act again. Some systems can keep this going for a long time.
So the difference between a model and a creature cannot be “one loops and the other does not.” That is false.
The difference is that the model's loop is arranged. It is started from outside, maintained by outside machinery, aimed at a task given from outside, and stopped by a condition that is not its own death. When the call ends, the model has not gone hungry, healed, panicked, slept, or conserved energy. It has not preserved itself through the interval. In the strict sense relevant here, nothing is waiting in the gap.
A living thing is different. It is never simply called and returned. Its body is already running before any question is asked. It is regulating temperature, chemistry, posture, blood sugar, pressure, damage, attention, and readiness. The world is not an input submitted to it. The world is already in contact with it.
A language model has a loop, but not a life.
4. The ticking body clock
The beetle thought Well what if it is just a deep, dark hole with nothing in it? Maybe that wouldn’t be so bad. Wouldn’t it be the easiest thing in the world to just crawl as deep as I can, tuck in my legs, and stay there?
Nothing would ever go wrong in the little hole. The beetle would never be surprised by anything, or have to make any decisions, ever again.
Then the beetle’s tummy rumbled.
The beetle's rumbling stomach is not a suggestion. It is not a user request. It is not a message that can be ignored forever while the system waits in a clean idle state. It is the body reporting that the conditions for continued beetleness are changing.
That is the standing prompt.
A living predictor is continually being asked, by its own body and by the world that body is coupled to, “What now?” The question does not arrive as language. It arrives as hunger, thirst, pain, warmth, cold, pressure, balance, fatigue, arousal, startle, curiosity, and unease. It arrives as the changing shape of the scene. It arrives as the fact that doing nothing is still an action with consequences.
The beetle can imagine a place with no surprises, but the fantasy lasts only until the body speaks. The dark room is not a solution because the beetle is not trying to minimise sensation in the abstract. It is trying to remain the kind of organised process that can go on being a beetle. Silence does not feed it. Stillness does not repair it. A perfectly unsurprising tomb is still a tomb.