Scaling-law papers showed predictable relations among loss, parameters, data, and FLOPs, guiding LLM investment.
Kaplan et al. (OpenAI, 2020) and later Hoffmann et al. (Chinchilla, DeepMind, 2022) quantified how loss improves when scaling model, data, and compute.
Those curves justified multi-billion-dollar clusters and “train bigger” versus “train more tokens” decisions. They became board-and-lab language.
The laws are not eternal physics, but they are empirical compasses that organized the 2020–2026 race.
World-model research sought dynamics prediction for planning—beyond pretty pixel generation.
DeepMind beats classical baselines in scientific prediction with a new deep learning system.
The 2026 debate on opaque reasoning vs. auditable CoT sharpens with Astra: Ryan Greenblatt (Redwood) and UK AISI concerns ask whether we can monitor what the model never writes.