Intellect-3: A 100B+ MoE trained with large-scale RL | Not Hacker News!