Full talk · edited transcript

You Need LLM-as-a-Judge! How to Stop Evaluating by Eye and Set Up Validation

Alisa Glushko23 May 2026

Alisa Glushko, AGI NLP team, SBER.

The talk covers: problems of evaluating neural network models; what an LLM-as-a-Judge is and why to train one instead of just writing a prompt; the judge Pollux (text); usage scenarios (agents, RAG, monitoring, reward function); and the new project Pollux Vision for evaluating video generation.

Continue reading — it's free

Google or a one-time email code. No password.