OpenAI's 3.4-month compute doubling trend used ~15 data points; Epoch AI corrected it to ~6
ohlennart · x · 2026-11-04
OpenAI's famous "3.4-month compute doubling" trend was based on only 15 data points (the chart shows a 300,000x increase from AlexNet to AlphaGo Zero). A group of researchers later corrected it to 6 months using 100 data points — that group became Epoch AI. The quoted post adds that Gordon Moore based his original prediction on just 5 data points.
More from Infra
- Formalizing PTX in Lean to formally verify GPU kernels written by AI agents — gharik · 2026-11-04
- fal Raises $1.65B in Combined Series E and F Rounds, NVIDIA Joins — isidentical · 2026-11-04
- Baseten launches Grid, Sandboxes and Loops in one week to unify inference and training — baseten · 2026-11-04
- Meta seeks $25B in bonds to fund AI buildout after S&P downgrade — ns123abc · 2026-11-04
- Snowflake launches Cortex Training: serverless fine-tuning for open-weight LLMs, claims up to 61% cost savings — StasBekman · 2026-11-04
- RISC-V Day Tokyo 2026 unveils agenda with Google TPU v3 on display and 2nm AI accelerator talks — DavidBennett__ · 2026-11-04