In 2017, Google shifted their message to investors from a company that organizes that worlds information to an “AI first” company. It was such a strong message from Google that at the time we decided to measure the frequency the company talked about AI on their earnings calls in 2017 and 2018. During that period it was common for company management to mention some form of AI or machine learning 20 plus times on each of their quarterly earnings calls. In 2019 through most of 2022, the frequency of Google mentioning AI marginally decreased and investors largely forgot that the company’s North Star was becoming an AI-first company.

When OpenAI jumped into our lives late last year, it restarted the investor question: how much progress has Google made in terms of AI? The answer for the past year has been Google’s AI initiatives have fallen short of the bar that OpenAI has set. Google’s previous foundation model, PALM, lacked the conversational fit and finish that has made ChatGPT mainstream. As a result, for the the past year Google investors have been concerned, does the company have the technical chops to be a leader in AI and what are the ramifications to the Search business?

As investors in GOOG, Deepwater has been closely watching what progress the company’s been making and proving with its generative AI offerings.

In early December, Google surprised investors by publishing a series of videos highlighting what Gemini will be able to do. In a nutshell, they did it. The “it” was building what appears to be the most advanced multimodal generative model.

Taking a step back, OpenAI’s runaway success has been built on a model that predominately uses text-based prompts that guide the chatbot to answer a question. This video makes it easy to understand the power of Gemini’s multimodal interface. It’s worth noting, Google edited the demo video, which reduced latency as well as editing out a number of prompts. The bottom line, Gemini does not run quite as smooth as the demo video would suggest. This reality doesn’t concern me because those bumps in the product will be smoothed out over the next couple years.

A little bit more on the importance of a multimodal interface. That means the model can act more like a human, taking in prompts from text, voice, and visual queues including photos, videos, and live-camera streams. This allows Gemini to better understand what the conversation is about, give more accurate insights, as well as pivot to unrelated topics. In short, it behaves more like a human. Gemini is not Artificial General Intelligence (AGI), however, when it’s released next year it will be the closest that we see to AGI.

I suspect by the time Gemini Ultra is released next year, GPT-5 will soon come out. That means that both Google and OpenAI will eventually have advanced multimodal chatbots. When it comes to building towards AGI, it’s less important to gauge which model has the most features in a given month, and more important to identify which companies have the capacity to play in the endless game of foundation model leapfrog.

Many investors believe Google is in a tight spot when it comes to capitalizing on generative AI. The thinking follows that generative AI will allow search-like features that compete with Google Search, to be embedded across the internet. That means most people won’t use Google Search as much. This of course would be disastrous for Google’s business model given 55% plus of their revenue comes from Search. On top of that, about 90% of Google’s business is somehow related to advertising (including Search, YouTube, Ad Exchange, and others).

I believe the Search experience we know well is going to meaningful change over the next year and generative AI will be a long term positive for Google’s Search business.

CEO Sundar Pichai’s message on the June earnings call outlined the company’s leadership in AI and their optimism about the “evolution of Search.” The company’s Search Generative Experience (SGE) that was previewed back in May will “make Search even more natural and intuitive.” SGE will be powered with a portfolio of AI products, including Gemini.

SGE’s engagement superpower is to increase the frequency people use Google by going beyond what Google Search gives users today. Pichai has outlined the new Search experience will “weigh multiple factors and personal preferences before making a purchase or booking a trip . . . that allow users to go deeper to learn about a topic.” This is the concept of personalized AI, something that Google should capitalize on with their 20 plus years of search data and understanding of how we think, buy, and where we go.

The new search results paradigm will have a greater distinction between types of search queries. Information queries (who is Abraham Lincoln?) will look the most different with a tiles-based approach. Commerce queries (coffee shops near me) will be a mix between tiles and some buy links. Navigation queries will likely change the least. At the bottom of the results will be buttons to go deeper on a topic or activity with the goal of keeping you in Search.

While the frequency of the standard blue links will decline, the probability the company will grow faster for longer should increase.

Disclaimer