Séb Krier, Google DeepMind AGI policy lead, requests research clarifying how LLM pre-training loss predicts scaling behavior · Digg