OpenAI

Wei and team don't directly offer any hypothesis about why Deep Research fails almost half the time, but the implicit answer is in the scaling of its ability with more compute. As they run more parallel tasks, and ask the model to evaluate multiple answers, the accuracy scales...