Generative AI Fundamentals Map

Map/Algorithms and learning/Wave 1

Algorithmic efficiency

The compute needed for a given capability keeps falling.

capabilityBuilt today Open on the map →

Facts

Lane: Algorithms and learning
Wave: 1 of 12 (build order, not a date)
Needs: 0 boxes all the way back through hard links
Unlocks: 7 boxes downstream through hard links
Leverage: None: already built
Cluster: Around datacentre capex

Evidence as of October 2026

Epoch AI found the compute needed to reach a given language-model performance halves roughly every eight months, with a 95% range of 5 to 14 months.

01Epoch AI: Algorithmic progress in language models

Around it

Needs · 0

Nothing beyond the detail list. It can start at any time.

Unlocks · 4

Capability price deflationcapability · Built today
Intelligence explosionstage · Speculative
Token demand explosionstage · Built today
Pretraining at scalecapability · Built today · helps, not required

Loops and off-ramps

AI research automationreinforcing loop
Autoresearchreinforcing loop
Commoditisationbalancing loop