Anthropic reportedly achieves ~25x year-over-year pretraining compute efficiency gains
sytelus · x · 2026-11-14
sytelus cautions against multiplying flops by a multiplier to estimate training compute. His takeaway from the plot: Anthropic has achieved roughly 25x year-over-year pretraining compute efficiency gains, though the polynomial nature of the curve suggests the headline multiple should be discounted.
More from Infra
- Kosmo and Ropedia release Physical AI dataset: 1,589 3D-scanned environments, 131 hours of egocentric video — liuziwei7 · 2026-11-15
- Retired Microsoft engineer explains what broke in the Outlook outage — nicolascraske · 2026-11-15
- AMD's Rumored Siren Halo APU Reportedly Packs Dedicated GDDR7 On-Package Memory, Eyeing Late 2028 — Terminator857 · 2026-11-15
- GLM-5.3 on Two DGX Stations Peaks at 607.8 tok/s but 32 Concurrency Causes 453s Latency — petrusenko_max · 2026-11-15
- Pundit likens possible Huang exit to Microsoft antitrust moment as rivals circle Nvidia — TiernanRayTech · 2026-11-15
- Analyst warns of hyperscaler monopoly: NVIDIA loses pricing power, startup funding dries up — BenBajarin · 2026-11-15