Human-seeded Evals: Scaling Judgement with LLMs (with Samuel Colvin) | Vanishing Gradients | Podwise