OpenAI research lead Noam Brown thinks certain AI ‘reasoning’ models could’ve arrived decades ago

Noam Brown, who led AI rational research in OpenAI, said some of the forms of “logic” AI models could reached 20 years ago “researchers were known” [the right] Method ”and algorithm.

“There were several reasons for neglecting the guidance of this study during a panel, Brown said during a panel.” Nvidia GTC Conference San Jose on Wednesday. “I noticed during my research that, okay, something is missing. People spend a lot of time to think before they act in a difficult situation. It will probably be very effective [in AI]The “

Carnegie Melon, along with Brown Pluribus, mentioned his work on Game-Playing AI at the University of Mellon, which defeated the elite human professionals of Poker. AI helped to create Brown that was unique in the sense that it was “rational” through problems without trying to get a more cruel-power system.

Brown is one of the architects behind 1 o 1, an OpenAI AI model that uses a technique Examination This is “thinking” before answering the question. Test-time infections include applying additional computing to running models to run a form of “logic”. In general, the so -called rational models are more accurate and reliable in domains, especially in mathematics and science than the traditional -based models.

During the panel, Brown was asked that academia could never expect to examine the scals of AI labs like OpenAI, which lacks access to the company’s computing resources. He acknowledges that in recent years it has become stronger because models have become more computing-intense, but educators can influence the fields that need less computing like model architecture design.

“[T]There is an opportunity here for cooperation between border labs [and academia]”Brown said. If there is this compulsory argument from the paper, you know, we’ll investigate it in these labs.”

Brown’s comments come at a time when Trump is creating administration Deeper Receiving scientific grants. AI experts, including Nobel laureate Jeffrey Hinton, have criticized these cuts, Say that they can threaten both domestic and abroad in AI research effortsThe

Brown called AI Benchmarking a region where academia could have an important impact. “The benchmark condition in AI is really bad, and it doesn’t require lots of counting to do it,” he said.

As we have written before, the popular AI benchmarks test today Mysterious knowledge Most people are at work that care. That led Wider Confusion About the ability and improvement of models.

4:06 PACIFIC updated: The previous version of this piece implies that Brown is referring to rational models like and 1 in its initial comments and 1. In fact, he was referring to his work in the game-play AI before his time in Open. We regret the error.