If Google wants you to know one thing, it’s this: Google has generative AI covered.
In the past few weeks, the company has made a slew of announcements and unveilings surrounding its generative AI portfolio. And while some assume the company to be a bit behind the eight-ball due to Microsoft and ChatGPT’s initial jump into the generative AI market, it’s clear Google has a comprehensive enterprise-grade approach. The following is everything you need to know about Google’s recent stake claim in the generative AI race.
Getting To Know Google’s Generative AI
First things first: Google is the only company on the planet that has commercial offerings that cover the entire AI stack – AI chips/compute, AI development tools/platforms and AI applications. It has developed its own generative AI “stack,” which is the basis of its current generative AI offerings. Google’s Tensor Processing Unit (TPU) chips are specifically designed for machine learning tasks and to be used with Google TensorFlow. Google’s TPUs are rivaling GPU and CPU units. Google claims TPUs are faster and more efficient at running AI workloads. In the next layer, Google offers a range of cloud compute options for AI workloads, including the managed service AI Platform. While these pieces aren’t Generative AI per se, they are critical AI foundation pieces designed specifically for all AI workloads, including Generative AI.
Now, back to the Generative AI stack.The stack is based on four foundational models. These include:
● PaLM 2: PaLM 2 is a large language model that’s been trained on more than 100 languages to complete tasks like text processing, classification, sentiment analysis, etc. PaLM 2 can “understand, generate, and translate” text, including nuanced text like riddles and poems. It’s the true foundation of Google’s generative AI models. (NOTE: PaLM2 is only one of several Google LLMs available for commercial use. Others include the legendary BERT, ELMO, RoBERTa, LaMDA and the most recent transformer LLM – T5)
● Codey: Codey is all about helping developers. It can be embedded in a standard development kit or application to improve the productivity of the developer by auto-generating code, auto-completing code and more. Codey is the guard rail to keep developers moving forward.
● Imagen: Imagen is a text-to-image model that allows users to create custom, studio-grade images. Imagen can also be used for image editing.
● Chirp: Chirp is trained for speech-to-text conversion. It was trained with a different architecture than Google’s other speech models. A single model can unifies data from multiple different languages, but users can specify what language they want the model to recognize.
Together, PaLM 2, Codey, Imagen and Chirp work to create Google’s generative AI solutions — including Bard, the Google chatbot competitor of ChapGPT. And if you weren’t paying attention, Bard is now available to everyone—even if you weren’t on the initial waitlist.
Google also announced Duet AI for Google Workspace Enterprise, which includes writing and content refinement in Gmail and Google Docs. It also allows users to create original images from text in Google Slides. Duet AI for Google Workspace is available for preorder, though not cost information is available.
Making Generative AI Accessible To All Users
Google clearly doesn’t want to roll its generative AI tools out piecemeal or to only certain user segments. Instead, Google’s Generative AI Studio is now available universally. Generative AI Studio lets developers who are not yet familiar with machine learning and AI build their own generative AI apps with text and images. The studio is based on Vertex AI, an ML platform as a service in Google Cloud. Users more familiar with machine learning and generative AI can bypass Generative AI Studio and use the Vertex AI-based Model Garden, which lets users search and interact with more than 60 of Google’s foundation models. You can explore the models and APIs here.
According to Google, data governance and security features are built into the Vertex AI platform to ensure that data remains secure.
Meanwhile, there’s also Gen App Builder. This program allows developers—pro and non-pro—to use Google’s foundation models to create their own generative AI apps. The no-code tool has already been adopted by Mayo Clinic to improve clinical workflows, coordinate information and hopefully improve the outcomes of patients by unifying data from various sources, in different types of formats. In particular, the Enterprise Search in Gen App Builder will help users find information faster to respond to better patient needs. Importantly: Google says Gen App Builder is also HIPAA compliant.
Need Help With Generative AI? Google Has You Covered
In addition to the tech advancement associated with generative AI, Google is focusing on business development associated with AI, as well. Google Cloud just launched a new AI consulting service and other tools to help companies discover the value of generative AI. Google Cloud Consulting unit will provide free learning programs online targeted to a variety of audiences. On-demand learning and credential programs are also available. The consulting will focus on helping customers find trends in their data, summarize information, automate processes and create content personalized to their customers and company.
Google Generative AI: Too Much Too Soon?
As with anything related to AI and generative AI, the question needs to be asked: is it all too much too soon? With so many concerns about generative AI and AI in general, Google’s announcements beg the question: do we really need all of these advancements released now? Yes. So many of Google’s new launches seem to be aimed toward making generative AI accessible to anyone—even those with little machine learning or AI experience—decreasing the barrier to entry. In any case, it’s certain that this is only the beginning for Google’s stake in generative AI. Stay tuned for more unveilings.