Abstract page for arXiv paper 2611.08522: What Are LLM Agents Actually Learning on Low-Budget Molecular Optimization Benchmarks?