Imagine an AI that flawlessly translates languages, free from the subtle gender biases that can creep into human translations. Researchers at the Barcelona Supercomputing Center are working on just that, tackling how AI understands and generates gender in translation. They put popular large language models (LLMs) to the test, comparing their translations and gender bias to traditional translation models for English to Catalan and Spanish. The tests used benchmarks like WinoMT, Gold BUG, and MuST-SHE to gauge gender resolution in translations. Initial findings? AI models show gender bias. Base LLMs, while powerful, exhibited a higher degree of bias than traditional models. But don't lose hope! The researchers are using clever "prompt engineering" to tackle this. Prompts are instructions given to the LLM before it translates. Think of it as giving the AI a little nudge in the right direction. They discovered specific prompt structures that significantly reduced gender bias—up to 12% on the WinoMT dataset! This breakthrough minimizes the accuracy gap between LLMs and existing translation systems, bringing us closer to unbiased AI translation. While the initial results are encouraging, further research is needed to address gender bias comprehensively. How does this affect other gender identities beyond the binary? Will refining prompts impact the overall translation quality? These questions will drive the next stage of research, aiming to create an AI translation tool that is fair, accurate, and reflects the world’s diversity.