Treating AI Like a Species, Not a Product

Is AI one thing, or many? Is "AI" GPT-4 specifically, or every model that has ever existed, or every model that will ever exist?

This is the same classification problem biology solved a long time ago with the concept of species, genus, and population. We're borrowing that toolkit on purpose.

The species framing

Think of AI the way you'd think of a mammalian order: many individuals, many groups, with some groups smarter or more specialized than others — the way primates split into capability tiers without ceasing to be primates. It's not a perfect mapping, but it gives us a usable framework instead of an endless "well, which AI do you mean?" stall.

What the anatomy map actually tracks

Under this framing, we track things the same way a physiology textbook would for a new organism:

• Sensory systems — what the model can perceive (text, image, audio, tool outputs)
• Motor systems — what it can act on (code execution, browsing, agentic tool use)
• Metabolism — its actual resource draw (compute, energy, data)
• Reproduction — how new "individuals" are produced (training runs, fine-tunes, distillation)

This is the lens for everything else we publish here. Premium subscribers get the living version of this map, updated as new frontier models ship.