llmflation

The article argues that the rapid decrease in AI inference costs is driven by software optimizations rather than hardware improvements, and that open-weight models running on consumer GPUs are becoming increasingly competitive with frontier models.