Microsoft CTO Kevin Scott already has an idea about the next generation of AI models. He thinks they might be good enough to pass PhD qualifying exams - but he's careful not to overstate it.
Microsoft's Chief Technology Officer Kevin Scott says he's already seen a preview of the next generation of AI models after GPT-4. In a talk with LinkedIn founder Reid Hoffman, he described what they might be able to do.
"If you think about GPT-4 and like that whole generation of models, they can perform as well as high school students on things like the AP exams," Scott says. AP (Advanced Placement) exams are college-level tests for high school students in the United States.
Scott says early looks at the next wave of AI hint that future models could pass PhD qualifying exams. He specifically mentions breakthroughs in reasoning that "will happen" and models that can better remember past conversations and information.
But the Microsoft CTO is quick to tone down this claim: "It probably sounds like a bigger deal than it actually is," Scott adds.
Ultimately, it comes down to how the technology is used in the real world. The fact that a model like GPT-4 can get a perfect score on an AP biology exam might be impressive, but it doesn't yet prove that it's useful, Scott says.
For Scott, a big moment in the growth of AI is how much easier it's become to use lately. As the cost of entry drops, more and more people can use AI models to solve problems. Tasks that would have taken him months to code as a computer science graduate 20 years ago now take a high school student a Saturday morning, he says.
Of course, this fits in with Microsoft's main AI business, which is selling access to models on the Azure cloud. At Microsoft's Build 2024 developer event about two weeks ago, Kevin Scott showed a chart that promised a much better OpenAI AI model by the end of the year.
