THE SCOOP OpenAI was founded, in large part, as an AI safety countermeasure to DeepMind, the AI company acquired by Google in 2014. But lately, one of DeepMind’s founders, Mustafa Suleyman, has been doing the unthinkable: looking under the hood at OpenAI’s crown jewels — its secret algorithms behind foundation models like GPT-4, people familiar with the matter said. That’s because Suleyman is now head of AI efforts at Microsoft, which has intellectual property rights to OpenAI’s software as part of its multibillion-dollar investment in the company. His presence, though, has brought new attention to an unusual dynamic: Microsoft and OpenAI are inextricably linked; they are also competitors. Some people at Sam Altman’s firm have bristled at the awkwardness of the arrangement since Suleyman came on board, some of the people said. The relationship between engineers at Microsoft and OpenAI is partly by design. The software giant was instrumental in building the compute power necessary to train the world’s largest AI models that became ChatGPT. That required some engineers at Microsoft to have intimate knowledge about the way OpenAI’s algorithms worked. Technically, Microsoft has access to OpenAI products that have launched, according to people familiar with the deal, but not its top secret research projects. But practically, Microsoft often sees OpenAI products long before they are publicly available, because the two companies work together closely to bring the products to market at scale. Today, both companies need one another. But if they decide in the future to go it alone, the presence of Suleyman at Microsoft gives it an ace up its sleeve when it comes to developing AI models. As a key figure in the rise of the technology, he could help recruit AI talent to compete with OpenAI. The Wall Street Journal reported that Microsoft is already moving to sideline OpenAI, building an “in-house OpenAI competitor inside Microsoft.” Semafor could not confirm that any foundation models close to the scale of OpenAI’s GPT-4 were being developed today at Microsoft. Microsoft, OpenAI and Suleyman declined to comment.  Denis Balibouse/ReutersREED’S VIEW Based on conversations with people at OpenAI, I don’t think the company is too nervous about Microsoft or Suleyman getting a good look at its technology. Advances in artificial intelligence are moving so fast that whatever Microsoft gets to see today will be quickly obsolete. If the relationship does start to cool, OpenAI could hold back on cutting-edge research. Microsoft has rights to OpenAI products, but only when they are officially launched as products, and can’t look at experimental research far from the commercialization phase, according to people at both companies. With the Tour de France coming up, here’s a bike analogy. OpenAI and Microsoft are in a two-person breakaway far ahead of the peloton. By working together, they might be able to keep the lead. If either goes solo, they may both fall behind. Both companies have to think about their finish-line strategy, when they will have to ditch the other. Part of that move for Nadella is making sure that, say, five years from now, the company is capable of developing sufficiently advanced AI models that it doesn’t need anyone else’s IP. And for Altman, that may mean being able to train its own algorithm without the help of a company like Microsoft. Today, at least at the top echelons of each firm, the relationship between the two companies is healthy. I don’t believe Microsoft is training a massive foundation model on the scale of what it’s building for OpenAI. It wouldn’t make sense for the company to waste tens of billions on that effort when it can use OpenAI’s models anyway. So I wouldn’t say Microsoft has an OpenAI competitor inside its walls. It’s more like the company wants to keep those muscles fresh, just in case it needs them one day. | 
