Deficiencies in LLM Multi-Turn Information Acquisition for Abductive Reasoning

Testing via the Alien Abduction game reveals that LLMs struggle with active evidence acquisition during multi-turn reasoning. Models often fail to update hypotheses effectively or prematurely commit to conclusions before exhausting available evidence.