Wider or Deeper? Scaling LLM Inference-Time Compute with Adaptive Tree Search arxiv.org 3 points by vrm 3 days ago