Скачать или смотреть GDPval paper by OpenAI Can AI actually do real work?

Can AI actually do real work?
I recently came across one of the most interesting papers of the past few weeks, called GDPval. It’s a new study from OpenAI that looks at how AI performs on real, economically valuable tasks, not quizzes or logic puzzles, but the kind of work that actually contributes to GDP.
What makes this study different is that it uses real tasks from professionals in sectors like finance, real estate, government, healthcare and manufacturing. These are single, complex deliverables that real people spend hours producing.
The evaluation method is also smart. Instead of using automated scoring or abstract metrics, experts compared AI and human outputs without knowing which was which, and simply chose which one they would use. That’s as close as we can get to measuring genuine quality.
The results are eye-opening. The best models are already producing work that matches human experts in several areas, especially when given time to reason and proper quality checks. Not every model helps some older ones can slow people down once you include the time needed to review their work but the overall direction is clear.
This study shows how close AI is getting to expert-level performance on real professional work. It also offers a useful way for organisations to test where AI can genuinely add value instead of relying on hype.
If you’re interested in how AI might reshape productivity in practical terms, this is a study worth paying attention to.
hashtag#AI hashtag#Research hashtag#Productivity hashtag#FutureOfWork