Study Examines How Compacting Reasoning History Affects Model Performance and Agent Cost
A new arXiv paper analyzes the causal effects of compacting long reasoning traces on future generation quality and agent costs, addressing the expense of long reasoning in language models.
Read the full story at arXiv cs.CL ↗