Musk says Grok 4.8 will finish pretraining this week and enter reinforcement learning