Apparently it's best in class with just under 27% accuracy on expert level questions. But don't worry, because it's getting close to 73% accuracy on questions that are 'conceptually simple' for humans, and almost 79% accuracy for the easiest ones. https://openai.com/index/introducing-deep-research/
Have you used it yet? I realize it's not perfect, but sure seems like a helpful first step in digging deep. Here's a review I asked it to do this morning on effective teacher to student feedback. https://chatgpt.com/share/67c0b0df-c394-800f-885f-34e0dd244365
Low accuracy is a deal breaker for me on usefulness for research. What do you see as the use case for this? I looked over the transcript (thanks!), and the tool is producing mediocre work. Aren't there better resources to get info on best practices for feedback? Are you using it for citations?
My results were a mixed-bag. But it jumpstarted the entire research process in a new way. And since this is likely to keep improving, I plan to embrace this as a staring point.
What’s your initial prompt or request to get started? I tried it as well for explaining concepts but could not see much difference to 4o. But I guess I should approach it in a more sophisticated way …
My prompt was complicated. It was for research on a baseball history article I am writing, so I had several steps in my prompt. ChatGPT then asked for clarification on several points before it compiled the research.
Comments
Research, though, isn’t perfect.