LLM companies use CPU technique from 1990s to speed up inference

Original: Anthropic.

Source: x.com ↗