Microsoft hires DeepMind co-founder Mustafa Suleyman to run consumer AI

“We want to make sure that this next wave is one that for the consumer Microsoft can really, really create incredible products,” Suleyman said in an interview.

The new hires also mark another significant step by Microsoft to bolster its in-house AI capabilities and products, outside of the relationship with OpenAI. Last month, Microsoft invested $16 million in Mistral AI, a French rival to OpenAI. Nadella on Monday told OpenAI CEO Sam Altman about Suleyman and his team joining Microsoft, the company said.

“We congratulate Mustafa and Karén on their new roles and look forward to working with them,” an OpenAI spokesperson said in a statement. “Our partnership with Microsoft is grounded in leveraging Azure supercomputer infrastructure built for OpenAI to power our research and train our next-generation models—a core part of our mission. We will also continue developing useful AI products like ChatGPT aimed at widely distributing the benefits of AI and driving industry-wide innovation for everyone.”

Nadella said Microsoft is “very committed” to its partnership with OpenAI. Microsoft will continue to provide the startup with the cloud-computing power to build its AI models and will use those in its software. Suleyman will be tasked with forming them into compelling, well-designed products for consumers. “We have in the world a number of cutting-edge pre-trained models,” Suleyman said. “The real challenge we have is turning those models into actual products today.”

Inflection in June raised $1.3 billion in one of the largest funding rounds of Silicon Valley’s AI frenzy. That round valued the startup at $4 billion, a person familiar with the matter said at the time. Microsoft board member Reid Hoffman is a co-founder of the two-year-old startup, alongside Suleyman and Simonyan, and other investors include Microsoft co-founder Bill Gates, former Google CEO Eric Schmidt and Nvidia Corp.

In recent months, Inflection had talks with multiple investors about raising additional funds, according to people familiar with the conversations. The company declined to comment.

Google, after falling behind Microsoft in the generative AI race, is starting to catch up. In February, the company rolled out Gemini, its answer to OpenAI’s ChatGPT. The rollout hasn’t gone perfectly. Google pulled a tool designed to generate realistic-looking images of people after a flurry of criticism that its depictions of race were historically inaccurate. But earlier this week, Bloomberg reported  that the company was in talks to license Gemini to Apple Inc. for new iPhone features expected later this year. A deal with Apple would be Google’s highest-profile partnership for Gemini to date, building on the two companies’ relationship around search and making it harder for Microsoft to make gains on smartphones.

Inflection’s Pi chatbot was designed to mimic human understanding of emotions and interact with users in a supportive fashion. While attracting considerable investor interest, including from Microsoft, and a million active daily users, the startup has not succeeded in finding an effective business model, Suleyman said. In the meantime, Nadella asked him to move over to Microsoft.

“We want to bring real competition,” Nadella said in the interview. “The thing that’s beautiful is when the medium changes, it means we get to play again to say, what’s the browser mean? What does even the operating system mean? What is an assistant? And so that’s the exciting part—not what happened but what is going to happen—it’s the rebirth of personal computing.”

Microsoft’s consumer AI products have had their own issues. The Bing chat initially got disturbingly personal with some users, forcing the company to temporarily cut conversations short. Users in February reported Copilot was generating responses that were called bizarre, disturbing and, in some cases, harmful. Microsoft said users deliberately manipulated Copilot to generate some of those responses. This month, a company engineer sent letters to the board, lawmakers and the Federal Trade Commission warning that Microsoft wasn’t doing enough to safeguard its AI image generation tool from creating abusive and violent content.

The work Inflection has done on tone, as well as the new hires’ experience in training AI software after it has initially been developed, will be critical here, Nadella said. “One of the key insights that Pi exemplifies is that pre-training is super-important and necessary but not sufficient,” he said. “All of us in AI, at least when it comes to product-making, are realizing that the post-training step may be everything.”

Meanwhile, Pi will remain available as a chatbot app, though that could change in the future as the startup transitions to a business-to-business model that licenses an API rather than a product, a person familiar with the company said.

DeepMind was founded in 2010 and acquired four years later by Google. Suleyman was a key public face for the company, speaking to officials and at events about the promise of AI and the ethical guardrails needed around the technology.

However, DeepMind was heavily criticized for its work in the UK health sector. Its first health product was an app called Streams that was originally designed to help doctors identify patients at risk of developing acute kidney injury. In July 2017, the UK’s data privacy watchdog said DeepMind’s partner in the project, London’s Royal Free Hospital, illegally gave the company access to 1.6 million patient records. Suleyman apologized in a statement at the time.

DeepMind workers complained about his management style, the Financial Times reported. Addressing the complaints at the time, Suleyman said: “I really screwed up. I was very demanding and pretty relentless.” He added that he set “pretty unreasonable expectations” that led to “a very rough environment for some people. I remain very sorry about the impact that caused people and the hurt that people felt there.”

Suleyman was placed on leave in 2019 and months later moved to Google, where he led AI product management until exiting in 2022.

In the interview Monday, Suleyman said he learned a lot from that period about the need for user trust and said Pi's design was heavily influenced by those lessons. “One of the things I earned was that trust is absolutely critical and being super transparent with users and repeatedly doing the same thing and being able to demonstrate that our products are not just reliable that users feel like they're in control.”

Nadella and Suleyman met years ago when Nadella gave a talk about his book Hit Refresh in London, and the two kept in touch. Conversations about joining Microsoft began in the past several months.