Scaling Law Extrapolation Analysis

via “scaling-law-extrapolation-analysis”

* ⭐ 06/2022: [Solving Quantitative Reasoning Problems with Language Models (Minerva)](https://arxiv.org/abs/2206.14858)

Unique: BIG-bench's scaling analysis is built on a diverse task set (204 tasks) rather than a single benchmark, allowing researchers to observe how different capability types scale differently — some tasks show smooth power-law scaling while others exhibit sudden emergence or saturation, providing richer insights than single-benchmark scaling studies