Researchers have been using large language models (LLMs) to design methods for complex AI tasks. They evaluated these agent-designed methods on various open-ended tasks and found that while they can occasionally match or surpass human performance, their designs are largely based on existing human-designed algorithms. In fact, nearly half of the agent-designed methods were identical to existing human-designed algorithms. This suggests that current LLM agents may not be creating entirely new approaches, but rather recombining existing ones. --- Why it matters: This research matters because it sheds light on the limitations and potential biases of AI systems that design other AI systems. Understanding how these systems operate can help developers create more transparent and reliable AI methods. Source: https://arxiv.org/abs/2608.17471