Kellblog

The Brute-Force Era of AI (and What Comes After)

The article argues that AI is currently in a brute‑force era where gains come from scaling data, compute and model size rather than new algorithms. While larger pre‑training corpora and longer context windows improve performance, the improvements are attributed to scale, not fundamental breakthroughs. Emerging techniques such as mixture‑of‑experts, sparsity and quantization allow trillion‑parameter models to run efficiently at inference time, hinting at a second phase focused on efficiency. The author predicts a two‑phase cycle: initial scaling followed by cost‑driven optimization, which will shape the long‑term economic value of AI.