Microsoft Puts the PR in AI (Premium)

Overly reliant on OpenAI and facing a coming generation of on-device AI that it can’t control, Microsoft’s latest small language model (SLM) is most notable for the PR offensive that accompanies it.

Phi-3 Mini is an SLM that runs on locally on smartphones and rivals the performance of the ChatGPT 3.5 and Mixtral 8x7B previous-generation large language models (LLMs) that run in cloud datacenters. This is interesting for the same reasons that any local SLM is interesting. But Phi-3 Mini enters a market crowded with capable competitors, many of which aren’t saddled by the software giant’s dominant monopolistic baggage and questionable OpenAI partnership. And so Microsoft has flexed its muscles in a way that few competitors can by engaging in an expansive PR campaign to spread the word, not just in the tech space, but with mainstream news too.

The message: Microsoft isn’t just a major player in AI, Microsoft is still leading this space.

Officially, Microsoft announced Phi-3 Mini to the world with a whimper, via an academic whitepaper that few will read and even fewer could possibly understand. But behind the scenes, Microsoft has shouted this otherwise innocuous news to the rooftops by reaching out fair and wide, not just to all the usual tech sites and blogs, but more notably to mainstream news publications like The New York Times and Reuters, to explain why Phi-3 Mini is notable and see whether they might promote this advance for them.

That they all took the bait is rather impressive. Each is essentially promoting Microsoft’s internal AI prowess for it at a time when the company is facing an uncertain future in a suddenly crowded field of AI advances that seem to arrive each week, and usually from other companies and organizations. Microsoft may have jumpstarted the AI era that OpenAI ChatGPT-4 created, but it seems to be getting swept up in it now, too, as the rest of the industry jumps onboard.

“The smallest Phi-3 model can fit on a smartphone, so it can be used even if it’s not connected to the internet,” The New York Times explains in its lengthy report. “And it can run on the kinds of chips that power regular computers, rather than more expensive processors made by Nvidia.” The publication also helpfully notes that Microsoft is open sourcing Phi-3, just as Google and Meta are doing with their SLMs.

None of that is particularly unique. But then again, that’s true of most of the new LLMs and SLMs that seem to appear almost every other day now. The claims are always the same: Better datasets, better training, better performance, and better capabilities, followed by rote comparisons with current and previous generation language models from the perceived market leaders (usually OpenAI). It’s all happening so fast that it’s difficult if not impossible to keep up with the advances. Which is exactly why Microsoft courted the press so gratuitously for this SLM. It would have otherwise been lost in the mix, ignored.

Microsoft cannot afford to be lost in the mix right now. It’s in the middle of a major push to realign its entire business around AI, a massive change that saw it bring in outsiders to lead a new Microsoft AI “organization,” whatever that means, a change that roiled AI leaders inside the company, leading to defections, one very public. This isn’t a pivot, it’s a coup.

The fear for Microsoft is that AI, for all its advances, won’t change the status quo. That it will, in effect, miss the next wave yet again. There is a very real chance that AI will simply extend the power and dominance of the companies that today dominate in search—and the advertising that accompanies it–and in smartphones and other mobile devices. And that is where the unique power of companies like Apple, Google, and Samsung comes into play. As they build more local AI capabilities into their devices and platforms, they can cement their leads in the foundational underpinnings of AI and ensure that this next wave is experienced via the same services and devices that dominate today. The same services and devices that relegated Windows and the PC to the backwaters of the personal technology industry.

Microsoft’s aggressive AI push is predicated on the belief that it could once again find itself on top. So far, it’s worked: Excited by the unexpected brashness of its strategy, Wall Street rewarded Microsoft by making it the biggest company in the world by market cap. But Microsoft’s position is also surprisingly tenuous. It has long been a distant number two in the cloud, and it has been a distant number three in device platforms, behind iOS and Android, for even longer. These businesses—Azure and Windows—are crucial to Microsoft, and they are financially successful. But the rest of the industry isn’t just stepping aside to allow a new era of Microsoft dominance.

SLMs like Phi-3 can work as a sort of end-run around the built-in AI capabilities that Apple, Google, and Samsung are building into their platforms. That is, if this or other third-party AI solutions are superior in whatever capabilities one gets from the platform maker, developers will adopt them for use in their own apps and services. And Microsoft, like Google, can take advantage of hybrid AI solutions that hand off to the cloud when the on-device capabilities fall short. This could still work.

But the software underbelly of this strategy is that AI will soon be so pervasive that it will be everywhere. And if free or cheap AI is good enough, then expensive, paid AI services will be less viable. A few examples may serve to make this point clearer.

The vast array of capabilities in Microsoft Office makes this family of apps and services a no-brainer for many companies. But with the democratization of AI and the rise of high-quality and free/inexpensive alternatives, that’s often no longer the case. Looking just at a simple example like spelling and grammar checking, the quality of free AI-based tools like Grammarly long ago eclipsed what Microsoft offers. And these tools work everywhere, turning otherwise non-competitive word processors, editors, and other apps into viable alternatives. I’m writing this article in a plain-text Markdown editor, for example, and while it does have its own spelling and grammar checking built-in, I’m also using the free version of Grammarly with it. Not only is Microsoft not required, it’s superfluous.

Now consider Copilot for Microsoft 365. This expensive add-on—in addition to a paid commercial license for Microsoft 365, businesses pay another $30 per user every month for this functionality—lives in a pane inside Microsoft Word (and other Office apps), helping you write, create new content, summarize documents, and the like. The differentiator here is that it’s built into Microsoft’s apps and runs against the Microsoft Graph. But any company could make Office add-ons that do the same, and these offerings could be less expensive and, like Grammarly, perhaps superior as well. There’s a price war coming.

On the backend, Microsoft’s unique relationship with OpenAI raises questions, yes, but the combination of ChatGPT’s technical prowess and the vast scope of Microsoft Azure is a powerful one-two punch. But it can be met by Google, unilaterally, and by hybrid solutions that combine any backend—Azure, AWS, Google Cloud—with any front-end built on any LLM or, more interestingly, any combination of LLMs. The competition is coming. And it’s coming fast: This past month alone has seen a blizzard of AI announcements from every corner. That won’t stop anytime soon.

With all that as context, the PR push behind Phi-3 is everything you need to know about how much the world has changed since Microsoft first introduced us to Copilot 15 months ago. Back then, it recast the public perception of AI beyond OpenAI and positioned itself as a leader. But today, this market is crowded, complicated, and moving fast, and it has already moved past OpenAI and Microsoft. It’s not clear which companies, which solutions, or which combinations of these things will emerge as winners. But it increasingly seems like this market will work much like AI itself, where there is no single killer app, but rather a crowded market of solutions, each of which speaks to different vertical markets.

Here’s the good news. Whichever companies come out ahead—and, yes, I do see Microsoft playing a big role at multiple levels in the stack—the ultimate winners will be us, the users. Let these behemoths and smaller players fight it out: We can simply choose the right tool—more likely, the right tools—for the job. It won’t always come from Microsoft.