Eternal City.
The Eternal City.
Six gates, a machine in thirteen parts, and a comparison with the brain.
FACT is a checkable record. CLAIM is an argument. FAITH is what the Order holds, and does not ask you to treat it as proof.
The charge.
Build the Eternal City: learn AI, build computers, wire networks, train models, and know how they differ from the brain.
The road.
Six arches, one for each gate. A packet asks (SYN). The far side agrees (SYN-ACK). The first side confirms (ACK). That is the handshake of a connection, not a sacrament.
The grid is a picture of many small arithmetic units. The dome below is a picture of a net, beside a picture of a cell. Neither is the brain.
The name.
Tibullus 2.5.23 speaks of the “eternal city.”
Source. Tibullus 2.5.23, “aeternae … urbis.” Cited as text.
Virgil, Aeneid 1.279: “imperium sine fine.”
Source. Virgil, Aeneid 1.279. Cited as text.
The work.
The eternal city, in this Order, is a city of builders. The poems name the hope. The benches, the wires, and the tests are the work. The same page is served at /silicon.
Silicon. Build a PC.
Assemble the thirteen parts. Know what each one is for before you buy a faster one.
Roads. Networks.
Addresses, routes, and the difference between a local wire and the public internet.
Scriptorium. Code.
Write a small program you can read a month later. Name things for the next reader.
Forum. Machine learning.
A model is a fitted function. It is not a witness. Check it against data it did not train on.
Pantheon. Neural networks.
Layers of weighted sums and simple nonlinearities. Inspired by brains. Not copies of them.
Capitol. Safety and law.
Say what the system must not do. Keep a log. A renamed technology is still the same technology.
Thirteen parts of a machine.
| Part | Name | What it is |
|---|---|---|
| 1 | Case | The box that holds the parts and keeps hands off the mains. |
| 2 | Power supply | Turns wall current into the voltages the parts use. Size it for the draw, with margin. |
| 3 | Motherboard | The roads between the parts. Slots, sockets, and the firmware chip. |
| 4 | Processor | Does the arithmetic. Cores are workers. Clock is how fast each step is offered, not how wise it is. |
| 5 | Memory | Short-term workspace. It forgets when the power goes. |
| 6 | Storage | The record that remains: disk or solid-state. |
| 7 | Graphics processor | Many small arithmetic units. Useful for pictures and for training. |
| 8 | Cooling | Heat is the tax. If you do not pay it, the machine slows or stops. |
| 9 | Network interface | The gate to other machines. Wired or wireless. |
| 10 | Input | Keyboard, pointer, sensor. How a person or a world gets in. |
| 11 | Display | How the machine answers the eye. Not required for a server. |
| 12 | Operating system | The steward of time, files, and permissions. |
| 13 | Firmware | The first program, on the board, that knows how to start the rest. |
Brain, by the numbers.
The human brain has on the order of 86 billion neurons, about 10¹⁴ to 10¹⁵ synapses, and a power draw near 20 watts.
Source. Herculano-Houzel, Frontiers in Human Neuroscience (2009), for the neuron count. Synapse and watt figures are the usual textbook range, not a measurement made here.
Neural net and brain.
| Brain | A common neural net |
|---|---|
| About 86 billion neurons | Units that compute a weighted sum and a simple function |
| About 10¹⁴ to 10¹⁵ synapses | Weights, stored as numbers |
| About 20 watts | Training runs that fill data centres, measured in megawatts |
| Learns while living | Usually trained by backpropagation on a set, then frozen or fine-tuned |
Inspired by, not copies of.
Backprop.
Whether brains use backpropagation is an open question. Do not teach it as a settled fact about neurons.
Today no mind-upload or life-extension procedure exists; the Order offers, sells and promises none.
Source. No such procedure is in clinical use. The Order states the limit itself.