descriptive claim
Algorithmic progress roughly halves the compute required to reach a fixed language-model performance threshold every ~8 months, so algorithmic efficiency contributes comparably to raw hardware scaling in observed capability gains.
desc_algorithmic_efficiency
confidence 0.70
Warrants (2)
supports (1)
- Algorithmic Progress in Language Models modeled_projectionweight0.85
locator: Abstract; headline finding
“compute required to reach a set performance threshold has halved approximately every 8 months”
qualifies (1)
- Epoch AI --- compute, data, and algorithmic trends expert_estimateweight0.50
locator: Epoch commentary on benchmark-dependence of efficiency estimates