Code clone detection is one of the fundamental problems in software maintenance. While traditional code clone detection techniques have achieved competitive results in identifying syntactic similarity, their ability to handle semantic and cross-language clones remains constrained. The emergence of large language models (LLMs) introduces new opportunities for exploring semantic relationships in source code, which at first glance appear suitable for clone detection tasks. In this paper, we present an empirical comparison between LLM-based clone detection and a representative traditional tool (LICCA) across both intra-language and cross-language scenarios. We analyze the respective strengths and limitations of the two approaches and examine the impact of prompt formulation on detection performance. The experimental results suggest that integrating traditional clone detection tools with LLM-based semantic analysis may help address complementary strengths and limitations observed across the evaluated approaches.