Cohere survey of scaling laws: 45 of 50+ papers use power laws, most under-report details
Cohere · youtube · 2026-09-19
A Cohere Labs talk by Sneha Kudugunta surveys scaling laws research: fitting details (equation, training setup, optimizer) can flip conclusions like the optimal token-to-parameter ratio. Of 50+ surveyed papers, 45 quantify trends with power laws, but most under-report details needed for reproduction. The team proposes an authoring checklist for scaling law research.
More from Research
- Anthropic sets up wet lab in Bay Area to pursue rare disease treatments — ivan_bezdomny · 2026-09-19
- GRF-Recon: Global Ray-Field Optimization Tackles Long-Sequence 3D Reconstruction — zhenjun_zhao · 2026-09-19
- Seeing Is Not Remembering: PersistBench Exposes 4D Models' Weak Visual Memory — zhenjun_zhao · 2026-09-19
- AMB3R-SLAM: Kilometer-Scale Monocular SLAM on One Consumer GPU, 70% Lower ATE — zhenjun_zhao · 2026-09-19
- EliGSiR: continual RGB-D Gaussian Splatting mapping under bounded compute hits 21.52 dB on TUM RGB-D — zhenjun_zhao · 2026-09-19
- VGGT-GS SLAM: uncalibrated monocular Gaussian Splatting SLAM built on feed-forward VGGT priors — zhenjun_zhao · 2026-09-19