Recitation over Reasoning: How Cutting-Edge Language Models Can Fail on Elementary School-Level Reasoning Problems?
This reveals a critical flaw in LLMs' reasoning capabilities, challenging claims of near-human intelligence and urging the community to re-evaluate benchmarks and model training.
The paper investigates whether cutting-edge language models possess true reasoning ability or merely recite training data by creating RoR-Bench, a multi-modal benchmark with subtly shifted conditions. It finds that models like OpenAI-o1 and DeepSeek-R1 suffer up to 60% performance loss on elementary school-level problems when conditions are slightly altered, indicating severe recitation behavior.
The rapid escalation from elementary school-level to frontier problems of the difficulty for LLM benchmarks in recent years have weaved a miracle for researchers that we are only inches away from surpassing human intelligence. However, is the LLMs' remarkable reasoning ability indeed comes from true intelligence by human standards, or are they simply reciting solutions witnessed during training at an Internet level? To study this problem, we propose RoR-Bench, a novel, multi-modal benchmark for detecting LLM's recitation behavior when asked simple reasoning problems but with conditions subtly shifted, and conduct empirical analysis on our benchmark. Surprisingly, we found existing cutting-edge LLMs unanimously exhibits extremely severe recitation behavior; by changing one phrase in the condition, top models such as OpenAI-o1 and DeepSeek-R1 can suffer 60 percent performance loss on elementary school-level arithmetic and reasoning problems. Such findings are a wake-up call to the LLM community that compels us to re-evaluate the true intelligence level of cutting-edge LLMs.