OpenAI's latest initiative explores how GPT-5 can accelerate scientific discovery. Collaborating with institutions like Vanderbilt, UC Berkeley, and Oxford, researchers documented early experiments across various fields, including biology, mathematics, and computer science. One notable example involved Dr. Derya Unutmaz, who used GPT-5 to quickly identify a mechanism behind a puzzling change in human immune cells, drastically reducing the time needed for hypothesis generation and experimentation. In mathematics, researchers Mehtaab Sawhney and Mark Sellke tackled a long-standing problem proposed by Paul Erdős. Stuck on the final step, they received a breakthrough idea from GPT-5 that helped them complete their proof. Similarly, Sébastien Bubeck and Christian Coester discovered flaws in a commonly used decision-making method in robotics with the model's assistance. These instances highlight GPT-5's potential to enhance understanding and efficiency in complex scientific inquiries. The collaboration approach is key. Scientists define the questions and methods while GPT-5 provides rapid exploration of ideas and connections. Researchers learn to engage with the model effectively, treating it as a dialogue to refine their hypotheses. While GPT-5 doesn't autonomously solve problems, it can streamline workflows and assist in literature searches, revealing relevant connections that might otherwise go unnoticed. Early results show that in structured fields like mathematics, the model can significantly reduce the time spent on developing viable proofs.