🎯 LLM-as-judge evaluation

The items that name it 2

• Cutting our inference bill by 68% by being obsessive about optimization
  writeup2026-10-19

  Two models from different providers labelled a blinded sample of 657 findings as important, minor, duplicate, wrong or unverifiable and rated severity. A third model settled disagreements.

• Falcon-Emirati: When an LLM Learns the Dialect, the Culture, and the Nuance
  article2026-10-06

In every workspace you can read 3