Join us in returning to NYC on June 5th to collaborate with executive leaders in exploring comprehensive methods for auditing AI models regarding bias, performance, and ethical compliance across diverse organizations. Find out how you can attend here.
In a cheeky nod to last year’s viral video featuring a montage of every utterance of the word “AI” during the Google I/O keynotes, CEO Sundar Pichai decided to turn the tables.
As he brought this year’s keynote to a close, Pichai couldn’t resist the temptation to showcase the capabilities of Google’s latest Gemini AI model. With a glimmer of amusement in his eyes, he fed the entire transcript of the keynote into Gemini’s eager algorithms, challenging it to count the number of times “AI” appeared. The result? A staggering 120 mentions, a testament to Google’s unwavering commitment to AI.
Over the past few days, the announcements from Google I/O were nothing short of impressive. Google is going head-to-head with OpenAI and Microsoft, integrating Gemini into a wide range of products.
What Google did at I/O
I attended this year’s I/O conference and saw far more than an electric DeLorean. From the introduction of Astra, a multimodal AI tool, to Gemini’s integration across search, photos, and personal assistants, Google is cementing itself as a strong competitor in the AI landscape.
VB Event
The AI Impact Tour: The AI Audit
Join us as we return to NYC on June 5th to engage with top executive leaders, delving into strategies for auditing AI models to ensure fairness, optimal performance, and ethical compliance across diverse organizations. Secure your attendance for this exclusive invite-only event.
The variety of Gemini’s applications—from the lightweight Gemini Nano to the powerful Gemini 1.5 Pro and Flash—demonstrates Google’s ambitious plans. With innovations like AI Agents, AI Teammate and the new Gemini app, Google aims to make AI a central part of everyday digital experiences, highlighting its strategy to outpace its rivals.
After the keynotes, as I sit through many labs and breakout sessions, I begin to experience a growing sense of cognitive dissonance.
A reminder of keynotes past
The world shimmers and fades away, transporting me back to a distant memory from another time and place, of another tech giant and its eccentric CEO. The year is 2006, and a middle-aged man stands on a brightly illuminated stage, wearing a corporate blue button-down shirt and pleated khaki pants.
This man is sweating profusely, to the point of absurdity, as if he’s a living SNL skit. Wide-eyed, he pounds his fist into the palm of his hand and begins to chant the word “developers” over and over again, as if trying to summon a demon from the ether. His name is Steve Ballmer, the CEO of Microsoft, and he’s whipping the crowd into a frenzy. The incantation works, and the audience begins to chant and clap along. Developers, developers, developers!
Suddenly, I’m back in the Google I/O lab session. Here they are, developers, a hundred or more of them, are sitting in front of me and tuned in. Some of them are struggling to follow along, however. A few confused coders timidly raise their hands to ask questions but quickly put them down again as the speaker marches on, leaving no room for stragglers in the AI revolution.
Then it hits me: Google has the wrong mantra. The AI war is being fought for the hearts and minds of developers, but Google is distracted by its consumer core.
Fighting the wrong war?
It is developers who have to build the apps, use the tools and ultimately decide the victors. Enterprise developers are where Microsoft has the greatest advantage, and they are the real reason Microsoft and OpenAI are already winning, even though Google had the lead in AI. Steve Ballmer had it right all along.
Microsoft, in collaboration with OpenAI, remains focused on delighting developers. OpenAI, while known for its consumer-facing products, is actually highly developer-focused, providing comprehensive APIs, extensive documentation, and robust support to make AI integration seamless.
Even Meta with its open-source Llama 3 and larger family of Llama models is attempting to position itself as developer-friendly, allowing anyone to take its AI and build whatever they want with it, free of charge.
This is not the first time Google has fumbled a lead. Developer traction has been the Achilles’ heel of Google Cloud, a marvelous infrastructure that is technically superior in many respects.
Yet, after over a decade and $45 billion invested, Google Cloud still trails behind with an 11% market share, versus Amazon Web Services at 31% and Microsoft Azure at 25%. Google had a three-year head start, but Microsoft now has almost three times the market share. The reason? Developers.
Google’s renewed developer push with AI
Will Google’s AI push finally break Microsoft’s grip on developers? I asked some of them at I/O.
As our lab came to a close, I approached one developer who I noticed was particularly attentive during this session on fine-tuning Gemma models. Was she impressed with what she saw?
“I think impressed is a strong word. I’m currently working with [large language models] LLMs in my research and I am curious to see if I can use Gemma for my specific research topic which will involve fine-tuning”, said Layla Bouzoubaa, a doctoral candidate at Drexel University. “I use OpenAI regularly and I think it’s been very straightforward. I am learning how to use Gemma within Google Colab, and to me it seemed a little bit more intimidating to approach from a developer’s perspective.”
Despite some challenges, Google has plenty of inroads with developers and significant opportunities. Indeed, the company even landed OpenAI’s former developer advocate Logan Kilpatrick, which Business Insider called a big win in the war for AI talent.
Bouzoubaa sees fine-tuning smaller language models (SLMs) as something that is quickly becoming essential. “I don’t have the hardware to host my own model. So, before, fine-tuning a large model wasn’t really a possibility.”
However, with smaller language models, Bouzoubaa sees the potential for getting better results with health data. “I know the models aren’t necessarily trained on the health data that I use, so being able to fine-tune is going to [enable more] AI applications”.
Just a few rows away, another developer is heads-down on her laptop when I interrupt her workflow. “I’m kind of a Google fan girl,” said April Johnson from Extensis, a 30-year-old software company focused on managing licensed fonts for teams. “Microsoft is doing amazing stuff, but I think Google is still ahead with the deeper problems to solve with AI.”
Johnson is also quite enthusiastic about learning using Google tools. “I use Google Colab all the time and I like the way you can easily set up prototyping. It’s so easy to just [prototype] on my computer and then just move it to the cloud.”
Near the last row of the room, a man with a beard that suggests he almost certainly has the root password hunches over a laptop with a thousand stickers from a hundred conferences like this one. He is Nick Bates from Cyber Drive, a startup that has created a smartphone specifically for children.
“What we’re using ML for right now is to check for patterns of abuse”, Bates says. “We are not creating our own [AI models] ourselves, but with Gemma being open we’re looking at ways to leverage it.”
Another developer seated nearby who wishes to remain anonymous echoes, “The problem is the big models are really big, so you need a big machine, and you need a lot of users so that it is profitable to actually host yourself. Many of the open source models are so big I don’t have the space to run them [on my laptop].”
Microsoft’s edge with developers
As I leave the Google I/O conference, I can’t help but feel that despite Google’s renewed push to win over developers with AI, Microsoft still has the upper hand. The Redmond giant’s deep roots in the enterprise, combined with its strong developer ecosystem and the strategic partnership with OpenAI, give it a significant advantage in the battle for the hearts and minds of developers.
Microsoft’s release of the Phi-3 family of small language models, particularly the Phi-3-mini, demonstrates it is capable of competing with SLMs as well. The model’s optimization for various platforms and its support for larger context windows make it an attractive option for developers looking to integrate AI into their projects.
Google, on the other hand, while technically impressive, still has some way to go in terms of making its AI tools more accessible and user-friendly for developers. The feedback from developers at I/O suggests that Google’s AI offerings can be intimidating to approach, especially compared to the more streamlined experience offered by Microsoft and OpenAI.
As the AI race intensifies, it’s becoming increasingly clear that the key to success lies not just in the quality of the AI models themselves, but in the ability to empower developers to harness their potential.
Microsoft, with its long-standing developer focus and strategic partnerships, seems poised to maintain its grip on this crucial audience. Google, despite its technical prowess, will need to double down on its efforts to simplify and streamline its AI offerings for developers if it hopes to catch up.
In the end, the winner of the AI race may well be determined by which company can most effectively rally developers to its cause. And right now, it looks like Microsoft is the one still chanting that word.