arXivSamuel Lewis-Lim, Xingwei Tan, Mario Sanger, Zhixue Zhao, Nikolaos AletrasFri, Oct 2, 2026, 9:04 AM PDT
score 15.1
Shorter AI reasoning still explains model decisions, study finds
Original: Efficient Reasoning Training Does Not Always Harm CoT Faithfulness and Monitorability
Source: arxiv.org ↗
Writing ELI5 summary…