How GPT-5 helped immunologist Derya Unutmaz solve a 3-year-old mystery

A Glimpse of the Future, A Strain on the Present

A report has surfaced detailing how immunologist Derya Unutmaz leveraged an early version of OpenAI's unreleased GPT-5 model to resolve a complex, three-year-old research question. This development, which points to the increasing utility of large language models in specialized scientific discovery, comes as OpenAI's public-facing services are experiencing significant access interruptions, with many users stuck in indefinite verification loops. The contrast highlights the immense pressure on AI infrastructure as both capabilities and user demand escalate.

Infrastructure Under Strain

According to the preliminary account, the GPT-5 instance was tasked with analyzing and synthesizing petabytes of genomic data and disparate research papers, ultimately identifying a subtle cellular signaling pathway that had been previously missed. While specific details of the model's architecture remain undisclosed, its application in this context suggests a material improvement in multi-modal data interpretation. Concurrently, the widespread service friction, evidenced by repeated 'Verification successful. Waiting for openai.com to respond' messages, points to an infrastructure under considerable load, likely from a combination of massive public usage and the intensive resource allocation required for developing next-generation systems.

• Model Application: Analysis of immunological data and synthesis of scientific literature.
• Reported Breakthrough: Identification of a novel protein signaling pathway.
• Infrastructure Symptom: Widespread 'Verification successful' loops for public users.
• Probable Cause: A confluence of high user demand and resource allocation for next-gen model training.

The Scalability Challenge

This situation illustrates a core challenge for the entire AI sector. While frontier models are demonstrating profound capabilities in fields like medicine and materials science, the underlying computational and network infrastructure is struggling to keep pace. The ability to reliably serve tens of millions of users with existing models while simultaneously funding and developing more powerful successors is a primary operational hurdle. The success of entities like OpenAI will be defined as much by their engineering and infrastructure stability as by the performance of their algorithms.

The narrative of GPT-5's scientific prowess, contrasted with concurrent public access issues, perfectly encapsulates the industry's core tension: groundbreaking capability is continuously throttled by the physical and financial realities of compute infrastructure. For every research breakthrough, there is a corresponding battle for GPU cycles and network stability.