The A.I. war has gone from zero to 100 in the course of a week. Microsoft has integrated an upgraded version of OpenAI’s buzzy ChatGPT technology into its search engine Bing and Edge browser. And Google’s new ChatGPT competitor Bard created buzz—and criticism—setting up a battle between powerhouse Google search and second-place Bing.
Google’s new A.I. chatbot Bard had a rocky start, as the tool made a mistake in a recorded demo this week by providing inaccurate information, something ChatGPT and Microsoft’s A.I.-powered Bing have also done, but which is not a good look for Google’s new generative A.I. rival—and which sent Google parent Alphabet’s stock tumbling. But according to Gaurav Gupta, a partner at Lightspeed Venture Partners focused on tech and A.I. startups, dwelling on one Bard mistake could be, well, a mistake. He says it’s hard to tell whether the blunder was a “one off” or a systemic issue, and that “all these models have quirks when they are first released,” including ChatGPT’s technology. They’ll improve the models over time, he predicts. And in the meantime, the big question is how the two A.I. models compare—and which might gain dominance among real users.
Both companies’ moves to integrate A.I. into search means it’s a “head to head battle,” Gupta told Fortune. But he thinks that one company has an edge over the other: “Assuming the technologies are equivalent, or very close, the company with the most distribution will be more likely to win. And in this case, it's very obvious” who has that: Google.
The technology underlying ChatGPT, the A.I. question-and-answer chatbot launched by startup OpenAI, which boasts a new $10 billion infusion from Microsoft, has dazzled investors, technologists, and everyday users over recent months as its eerily human-like Q&As have gone viral. Google’s Bard is an “experimental conversational AI service” powered by Google’s Language Model for Dialogue Applications, otherwise known as LaMDA.
Though not publicly available yet (Google has opened it up to select testers this week), Bard is designed to improve users’ searches by offering A.I.-generated answers as part of a results page while also offering the normal articles and links (like a typical Google search). According to a presentation in Paris this week, Google’s Bard will enable users to do things like plan road trips or debate buying an electric car.
The integration of OpenAI’s ChatGPT technology with Microsoft’s Bing, meanwhile, looks similar: Answering prompts from users in conversational language side-by-side with regular links. Both bots are trained on enormous amounts of data, while Bing is evidently using an updated ChatGPT technology that allows it to provide answers about real-time events (Microsoft called their new Bing A.I. integration “more powerful” than ChatGPT).
One current Google employee working on A.I. at the company told Fortune that A.I. project timelines have been accelerated and Google is “trying to respond super quickly.” Of Bard’s mishap during its demo, they said it is definitely a water-cooler topic among Googlers, akin to the chatter about the controversy on Twitter.
However since both companies have plenty of money to spend on developing their A.I. efforts, relatively comparable talent, and access to research, “they'll be in a constant race to release better and better models every six months” and may remain neck and neck, says Gupta. “So distribution will be the key here,” he argues. In that sense, Google clearly has the upper hand given that it’s so ubiquitous as a home page for search. He compares it to Apple Pay’s dominance versus Google Pay: Despite what he calls “equivalent offerings,” Apple Pay has far greater market share since it has widespread distribution with the iPhone and it is the default.
But key to that equation is also the sheer amount of data that Google has—in fact, “one could argue that Google has access to more data than Microsoft,” says Gupta. That could help Google train a superior model since they “have all the search queries of everyone who's ever searched. They know what data sources are more trusted, which ones are more popular, and they could incorporate that into their results”—including data they can pull from YouTube for video, Gupta points out (he does note, however, that Microsoft has an advantage with enterprises with their Teams product).
Meanwhile Gupta believes that since Google’s LaMDA is “one of the best models that's ever existed” it's going to be “equivalent in terms of an offering,” or “possibly better.”
Hedging their bets?
A notable, and perhaps curious, element to this battle is that Google is also spending money on apparent rivals in the A.I. space. Just last week, the tech titan announced a partnership with A.I. research lab Anthropic, which is building another ChatGPT-and-Bard-style bot named Claude. Anthropic was founded by former OpenAI researchers, and Google also invested $300 million into the startup.
In addition to the investment (Google reportedly got a 10% stake in the company), Google Cloud will be Anthropic’s “preferred cloud provider” as it develops its A.I. systems. Gupta (who is not invested in Anthropic) points out that there are new companies that need to train their models, and there’s demand for cloud infrastructure—like Google Cloud or Amazon’s AWS; but additionally, Google needs to protect its ad business from A.I.-powered rivals like the new Bing-ChatGPT-tech mashup. In that sense, Google’s moves to both invest in a ChatGPT-type model with Anthropic and build their own seem to work for two “separate businesses almost acting independently, and I'm sure there's someone tying it all together at the very top,” says Gupta.
But apart from Microsoft, who could be losing from Google’s A.I. moves? A host of startups.
In Gupta’s eyes, things could be grim. He points out that some startups are spending millions of dollars on their models assuming they’ll be able to monetize them eventually. Those models need to be updated “every six months, and every time, the model becomes more costly.” That could create a hamster wheel scenario for startups that raised money in the A.I. venture funding boom while the tech titans power on with their versions with “almost unlimited bankrolls.” Most of the companies Lightspeed’s Gupta is speaking with are trying to ensure their systems are compatible with both Google and Microsoft’s models, or even an open source model, which they believe is “imminent,” he says.
For one, Gupta doesn’t believe the startups will find product-market fit, and will “probably end up going bankrupt,” or, instead of building their own models, use a pre-existing A.I. model from Google or Microsoft, he suggests.
And some VCs predict the A.I. war will turn into a stalemate as these technologies become commoditized. Wesley Chan, a former early Google employee who founded projects like Google Analytics and Google Voice, who is now a venture investor at his firm FPV Ventures, recently told Fortune in an evidently prescient conversation that “Microsoft has the best one with OpenAI right now, and then all of the sudden, Google leapfrogs them a little bit, and then Amazon, or Netflix, or one of these other guys with outrageous amounts of money comes in as a dark horse and, surprise, you’re like, ‘Oh, okay, wow, I didn’t see that one coming.’ And then what are these startups to do who don’t have money?”
