The News: On December 6, Google announced the debut of Gemini, the company’s newest family of AI models. Here are the key details:
- Gemini 1.0 is three different models, Ultra, Pro and Nano:
- Ultra is the largest Google model ever; it is designed for highly complex tasks. The availability date is slated for early 2024.
- Pro “is our best model for scaling across a wide range of tasks.”
- Nano is designed for on-device AI. It is the only model Google shared size, the smallest large language model (LLM) yet. “We trained two versions of Nano, with 1.8 B (Nano-1) and 3.25 B (Nano-2) parameters, targeting low and high memory devices respectively.”
- Gemini Ultra outperformed GPT-4 in eight of eight performance benchmarks, including measures for reasoning, math, coding. and multimodal.
- Gemini is designed as natively multimodal. Multimodal reasoning capabilities can help make sense of complex written and visual information. “This makes [Gemini] uniquely skilled at uncovering knowledge that can be difficult to discern amid vast amounts of data.”
- “Gemini 1.0 was trained to recognize and understand text, images, audio, and more at the same time, so it better understands nuanced information and can answer questions relating to complicated topics. This makes it especially good at explaining reasoning in complex subjects like math and physics.”
- Gemini was trained on custom-designed AI accelerators “at scale … designed it to be our most reliable and scalable model to train, and our most efficient to serve …. On TPUs, Gemini runs significantly faster than earlier, smaller, and less-capable models.”
- According to Google DeepMind’s Gemini paper, the training workload was still huge. “Training Gemini Ultra used a large fleet of TPUv4 accelerators across multiple datacenters. This represents a significant increase in scale over our prior flagship model PaLM-2 …. TPUv4 accelerators are deployed in ‘SuperPods’ of 4,096 chips.” So “large fleet” of each SuperPod of 4,096 chips.
- Gemini Pro is being leveraged for Bard and is now available in Google AI Studio and Google Cloud Vertex AI.
- Gemini Nano will be built into the Pixel 8 Pro smartphone; Nano is being tested in an early access program for Android developers.
- There are plans to build Gemini into other Google products such as Search, Ads, Chrome, and Duet AI.
- Experiments with Gemini in Search via the Search Generative Experience have yielded 40% reduction in latency in English in the US, alongside improvements in quality.
Read the announcement on the launch of Google Gemini on the Google blog website.
Read Google DeepMind’s Gemini report on Google website.
Why the Launch of LLM Gemini Will Underpin Google Revenue
Analyst Take: Google’s Gemini family of AI models is impressive on paper, and if as good as promised, they could be the LLMs of choice for a wide range of AI app developers. But for a company that recorded more than $279 billion in revenue in 2022, 80% of which came from advertising, Google’s market drivers for Gemini are way beyond competing with LLMs and hyperscaler AI developer platforms. Here are our thoughts.
Understanding Google Market Drivers
In 2022, Google/Alphabet revenue was $279 billion, more than $224 billion came from various forms of advertising, most of it search advertising. In 2022, Google Cloud revenue was $26.2 billion. In first quarter (Q1) 2023, Google Cloud reported its first profit in more than a decade. None of these numbers are small change, but given the scale/ratio of these economics, would it make sense that Google is not always thinking about search?
AI-Powered Search
While a slew of follow-up announcements came from Google Cloud CEO Thomas Kurian at the Google Cloud Applied AI Summit on December 13, note that the original Gemini announcement on December 6 was from Google and Alphabet CEO Sundar Pichai and Google DeepMind CEO Demis Hassabis. The basic way we search has not changed significantly in years.
Earlier this year, I wrote about Google’s Search Generative Experience, Google’s initiative for experimenting with generative AI in search: “There has been a great deal of speculation about how generative AI will impact search, but to date, there have not been any significant breakthroughs. When ChatGPT exploded on the marketplace, Google, a longtime leader in AI research and use, was roundly criticized for its lack of generative AI applications. Was Google asleep at the wheel or does Google know something about generative AI search the rest of us do not know? The smart money says the company does indeed know what next-generation search will look like.”
In analyst one-on-ones at the Copilot launch event, Microsoft CEO Satya Nadella told us the company felt there was a significant opportunity in search for Microsoft leveraging generative AI and OpenAI intellectual property (IP). Gemini Ultra and Nano are going to be heavily leveraged to take Google search to the next level. Chrome now features search summaries and will mine not only text but also images and video to return search results.
On-Device AI Search
Around the world, most users search on smartphones, not laptops and PCs, so it was no mistake that Gemini Nano is a primary focus of the Gemini family of models. It was unclear if Google will charge for Gemini Nano, but my guess is that it will be free, but NOT open sourced (see below).
AI Stack
Interestingly, none of these bigger drivers detract from how Gemini, on paper, stacks up as additive to Google Cloud’s AI stack suite. On paper, it is a great addition. At the Applied AI Summit, Google laid out in detail how integrated Gemini is into the Google Cloud AI stack. Google AI Studio and Vertex AI have and are adding sophisticated and competitive tooling, integrations, and features that are on par with the other major players, Amazon Web Services (AWS), Microsoft, and IBM.
Not Open Source for a Big Reason
There is a lot of momentum around open source AI. In the case of Gemini, it is easy to see now why the models are proprietary and not free because the Gemini models will underpin so many other proprietary Google products.
Conclusion
There are a significant number of reasons Google will invest heavily in generative AI innovation and there is no reason to believe the primary driver of that innovation, particularly in AI model development, will be to make proprietary Google products better.
Disclosure: The Futurum Group is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of The Futurum Group as a whole.
Other Insights from The Futurum Group:
Google Cloud Next: Vertex AI Heats Up Developer Platform Competition
Google Cloud’s TPU v5e Accelerates the AI Compute War
Adults in the Generative AI Rumpus Room: Google, Mayfield, Context.ai
Image Credit: Google
Author Information
Mark comes to The Futurum Group from Omdia’s Artificial Intelligence practice, where his focus was on natural language and AI use cases.
Previously, Mark worked as a consultant and analyst providing custom and syndicated qualitative market analysis with an emphasis on mobile technology and identifying trends and opportunities for companies like Syniverse and ABI Research. He has been cited by international media outlets including CNBC, The Wall Street Journal, Bloomberg Businessweek, and CNET. Based in Tampa, Florida, Mark is a veteran market research analyst with 25 years of experience interpreting technology business and holds a Bachelor of Science from the University of Florida.
