Language models offer vast promise for automation, stepping into the realm of combinatorial optimization problems (COPs), which are notoriously NP-hard. In the paper ‘ReEvo: Large Language Models as Hyper-Heuristics with Reflective Evolution’, the authors present Language Hyper-Heuristics (LHHs), which employ LLMs to generate heuristics with minimal human intervention. The ReEvo framework applies a reflective, evolutionary approach to design, utilizing LLM inference, Internet-scale knowledge, and evolutionary search. Results showcase that ReEvo-generated heuristics not only outperform human designs but also do so rapidly. LHHs demonstrate remarkable efficiency, opening opportunities for tackling novel, real-world applications.

**Highlights: **

• Introduction of Language Hyper-Heuristics (LHHs) using LLMs.
• ReEvo framework for reflective, evolutionary heuristic design surpassing human capabilities.
• Successful surpassing of state-of-the-art human designs with ReEvo’s heuristics.

**Opinion: ** The implications of integrating LLMs like ReEvo in algorithm design can revolutionize various industries, reducing reliance on expert human intervention and enabling rapid, effective solutions to complex problems. This methodology’s ability to compete with, and even surpass, human-designed heuristics is a testament to the innovative use of AI in the field.