Google recently held its annual developer conference Google I/O 2023. Google is a large real estate owner with arguably more data than any other tech company in the world. This advantage cannot be overstated when it comes to training large language models (LLMs). In addition to having a strategic advantage for future development of LLMs with data, Google can offer advertisers instant ROI.

The primary announcements from the event were:

• Google drops the waitlist for Bard and announces new features.
• Google launches new Large Language Model, PaLM2
• Unveils its new AI-powered Search.
• Google Cloud announces new A3 supercomputer VMs built to power LLMs.

Google drops the waitlist for Bard and announces new features

Among the more exciting announcements at Google I/O, the company dropped the waitlist for Bard and the chatbot is now available in 180 countries and territories. Bard supports English, Japanese, & Korean languages, and will soon support more than 40 languages. Google is also rolling out features such as better source citations, the ability to export content generated in Gmail and Google Docs, support for more visuals and an upcoming Google Lens integration to analyze pictures and write captions.

Background on Google’s Bard:

Earlier this year, Google’s stock (Alphabet) tumbled 7% when chatbot Bard was unable to complete a search with 100% accuracy. During the demonstration, Bard returned incorrect information about which telescope was the first to take pictures of a planet outside the Earth’s solar system. This was a minor mistake given how far large language models and generative AI has come, rather it was the timing that was a bit flawed as OpenAI’s ChatGPT, the chatbot powering competitor Microsoft Bing, had been dominating headlines since its November 30th launch.

Microsoft, being an opportunist, took it a step further and announced Bing would now be powered by a faster and more accurate version of GPT-3.5 one day after Bard’s failed demonstration: “We’re excited to announce the new Bing is running on a new, next-generation OpenAI large language model that is more powerful than ChatGPT and customized specifically for search. It takes key learnings and advancements from ChatGPT and GPT-3.5 – and it is even faster, more accurate and more capable.”

Both companies have been preparing for this moment for many years. Microsoft invested $1 billion into OpenAI a few years ago with a new $10 billion round announced last month. Meanwhile, Google acquired DeepMind in 2014. Google also previously developed conversational neural language models such as LaMDA, which is used by Google’s Bard for its conversational AI technology.

Despite the mishap with Bard, it would be a human-generated mistake to think Alphabet does not command a place of leadership right now in generative AI. Alphabet was one of the first tech companies to focus and invest on AI and natural language processing (NLP). We pointed out to our premium research members in July of 2022 that ChatGPT is based on transformer architecture that Google initially introduced in 2017 when we said:

“Transformers are becoming one of the most popular neural-network models by applying self-attention to detect how data elements in a series influence and depend on one another.

Sequential text, images and video data are used for self-supervised learning and pattern recognition, which results in more data being used to create better models. Prior to transformer models, labeled datasets had to be used to train neural networks.

Transformer models eliminate this need by finding patterns between elements mathematically, which substantially opens up what datasets can be used and how quickly.

Google first introduced transformer models in 2017 and transformers are used in Google and Bing Search. Transformers also led to BERT models, which stands for Bidirectional Encoder Representations from Transformers, and is commonly used for text sequences. Transformers are also used in GPT-3 (it’s the T in GPT) which improved from 1.5 billion parameters to 175 billion parameters. GPT-3 has the ability to report on queries it has not been specifically trained on.”

Earlier this month, Google’s CEO, Sundar Pichai, gently reminded the AI community of how cutting edge Google’s research is when he stated, “Transformer research project and our field-defining paper in 2017, as well as our important advances in diffusion models, are now the basis of many of the generative AI applications you're starting to see today.”

BERT was designed to help Google better understand search intent, as despite billions of searches every day, about 15% of those searches are for brand new terms. This prompted Google engineers to develop a model that could self-learn.

The result is that searches results are more accurate by taking into consideration the nuances of language.

Google launches new Large Language Model, PaLM2

Google launched a new Large Language Model, PaLM2, that will power the updated Bard AI chat tool and more than 25 other new products & features including productivity software (Gmail, Google Docs), Healthcare and Security.  

PaLM 2 has the following capabilities:

• Multilingual: The LLM is trained on more than 100 languages, which increases language proficiency
• Reasoning: The LLM’s dataset has improved logic, common sense reasoning and mathematics
• Coding: The LLM can generate code including programing languages such as Python, JavaScript and specialized languages such as Prolog, Fortran and Verilog.

Google Unveils its new AI-powered Search

The company has unveiled its new generative AI-powered search that will be subject to a waitlist. Google cites the example of the following search “what's better for a family with kids under 3 and a dog, bryce canyon or arches.” Previously, you had to break the question down into smaller ones, sort through the vast information available, and then put things together yourself. Now with generative AI, search will be able to better understand the question.

Generative AI will also provide a better experience for online shopping by instantly getting relevant information like reviews, images, and ratings. The new shopping experience is based on Google’s Shopping Graph, which has more than 35 billion product listings.

The company announced the ‘About this image’ feature, allowing users to identify fake images. It mentioned in its press release, “When the image and similar images were first indexed by Google, Where it may have first appeared, and Where else it’s been seen online (like on news, social, or fact checking sites)”.

Google launches new Large Language Model, PaLM2

The company launched the new Large Language Model, PaLM2, that will power the updated Bard AI chat tool and more than 25 other new products & features announced during the Google I/O 2023.

Its predecessor PaLM, launched in April 2022, was a 540 billion based parameter, and the company did not provide this detail for PaLM2. PaLM stands for Pathways Language Model. “What we found in our work is that it’s not really the sort of size of model — that the larger is not always better,” DeepMind VP Zoubin Ghahramani said in a press briefing ahead of the announcement. “That’s why we’ve provided a family of models of different sizes. We think that actually parameter count is not really a useful way of thinking about the capabilities of models and capabilities are really to be judged by people using the models and finding out whether they’re useful in the tests that they try to achieve with these models.”

PaLM2 is faster and more efficient than previous models. Some of the improvements highlighted by the company are that PaLM2 is trained for improved multilingual text, spanning over 100 languages, reasoning, and coding, including popular languages like Python & JavaScript. For example, due to the multilingual capabilities of PaLM2, it has helped Bard to expand to new languages. PaLM2 is available in four sizes: Gecko, the smallest, followed by Otter, Bison, and Unicorn. Other use cases include improved Workspace features while working in Gmail, Google Docs, and Google Sheets. PaLM2 can also be used for enterprise use cases like Med-PaLM2 in medical research and Sec-PaLM in cybersecurity.

The company also said that it’s working on a more powerful model called Gemini and it will also be available in various sizes so that it can be easily deployed to various products.

Google Cloud announces new A3 supercomputer VMs built to power LLMs

Google Cloud announced the A3 GPU supercomputer that can be used to train and run Artificial Intelligence and Machine Learning models. While the A3 GPU supercomputer is on a private preview waitlist, the previously announced G2 VMs are now in general availability. The G2 VMs are powered by the new Nvidia L4 Tensor Core GPUs. The company said that it is the first cloud provider to offer these new GPUs for serving generative AI workloads.

The A3 GPU VMs are made of eight Nvidia H100 Hopper architecture GPUs, 3.6 TB/s bisectional bandwidth between A3’s 8 GPUs via the Nvidia NVSwitch and NVLink 4.0, 4th Gen Intel Xeon Scalable processors, and 2TB of host memory.

The A3 supercomputer can deliver up to 26 exaFlops of AI performance, thereby improving the time and cost of training large machine learning models. The A3 workloads will be run on Google’s Jupiter data center networking fabric which the company states “scales to tens of thousands of highly interconnected GPUs and allows for full-bandwidth reconfigurable optical links that can adjust the topology on demand.”

Conclusion:

I would not be surprised if we exit 2023 with a reimagined way to use Search Engines. The iteration cycle here is likely to move quickly compared to AVs or the Metaverse, as there are real-world applications where AI can be applied without safety issues (AVs) or friction in terms of user adoption (Metaverse/VR headsets). Instead, the scale has already been built with Search being a viral, daily activity used by nearly every human on earth. AI advancements will simply improve what is already in place.

Cutting-edge chatbots can be quickly deployed on the search engines that already exist, and this is a substantial difference from other overhyped, early-stage technologies. Their accuracy may still need time, but they're probably not too far off from being deemed “reliable enough.”

Investors should expect that AI will become a winner(s)-take-all market. In time, the difference in how search and other applications operate in terms of user experience plus ROI for advertisers will help carve a larger lead.

Premium Members should check the forum for updates on our timing for an entry into the stock.

Recommended Reading:

Google Stock: Search Is On The Precipice Of Multi-Decade Disruption
Google Faces Biggest Lawsuit in Company History — What Companies Could Benefit