Google Cloud Announces New Foundation Models & Embeddings in Vertex AI

Generative AI is unstoppable. The last few months have been exciting with tech giants like OpenAI, Microsoft, and Salesforce unveiling some cutting-edge digital assistants, software, content creation tools, and apps. These smart helpers have empowered multiple businesses and brands to generate better ideas, simplify research, and repurpose content. They have completely transformed the way content is created, apps are built, and websites are designed.

Google’s new offering is yet another affirmative step in this direction. Google announced PaLM 2, their next-generation language model that boasts of improved ‘multilingual, reasoning, and coding capabilities’. Google has always been fixated on the idea of empowering its customers and developers with transformational foundation models to embrace innovation in different ways.

Recommended: Tableau + GPT: Ushering into a New Era of AI-led Business Analytics

Gen App Builder and Vertex AI – The Big Announcements

In March, Google announced: Gen App Builder and Vertex AI.

Gen App Builder allows even a developer with zero machine learning experience to create search apps and generative chat while the generative AI support in Vertex AI helps to expand the machine learning development platform that can access both foundation models and APIs. It also offers a variety of tools to customize and test with other models in Generative AI Studio.

New Models in Vertex AI

At Google I/O 2023, Google made some exciting announcements that further enhanced the offerings where customers can access new generative modalities and increased model exploitation and tuning options.

In the three new foundation models available in Vertex AI, users can:

• Access via API.
• Tune through a simple UI in Generative AI Studio.
• Deploy to a data science notebook.

Code Generation with Codey

It is a text-to-code foundation model that can be easily integrated into an SDK or application. It helps to enhance both code quality and developer productivity in terms of code generation and completion. Codey speeds up software development with the help of real-time code completion as well as generation. It can also customize according to a customer’s own codebase.

Features:

• It supports 20+ coding languages, including Go, Google Standard SQL, Java, Javascript, Python, and Typescript.
• Based on the context of the code typed into the prompt, Codey offers the following few lines.
• It produces code in response to developer prompts in natural language.
• Programmers can communicate with a bot to ask questions about debugging, documentation, picking up new ideas, and other aspects of programming.

Recommended: Unlocking Generative AI for Marketing and Sales 

Mask-Free Editing with Imagen

With this text-to-image foundation model, businesses and users can create and edit studio-quality photographs at scale with low latency and enterprise-grade data governance for any business purpose. Organizations can use Vertex AI to personalize and customize Imagen to their own business requirements by creating images using content that they own, such as current goods or brands.

Features:

• Image captioning, mask-free editing, and image upscaling, across over 300 languages.
• Classify and caption the image with the perfect description.
• Easy editing like changing an object’s color.
• Any image produced by Vertex AI is customer data, and the company is free to use it for marketing purposes.
• Upload existing images using Vertex AI’s built-in security and governance controls to protect data.
• Generated photos can be enhanced with captions and metadata, upsized to the necessary resolution, and iterated indefinitely.

Inclusivity with Chirp

The speech-to-text foundation model works seamlessly while interacting with customers in their native language. With captioning and voice assistance, it enables businesses to interact with customers in their own languages. Chirp applies the power of large models to speech tasks including voice control to captioning to voice help, whether it’s interacting with customer service virtual representatives in Spanish, captioning films spoken in Xhosa, or providing voice assistance in Balinese.

Features:

• In languages with fewer than 10 million speakers, relative accuracy has improved by up to 300%, with a 98% accuracy rate for English.
• Trained on millions of hours of audio.
• Offers the same quality (as the most frequently spoken languages in the world) to millions of new languages and dialects.

Create Sophisticated Applications Using Embeddings API 

For longer inputs, like texts with thousands of tokens, embeddings transform text and image data into multiple dimensions numerical vectors which represent semantic relationships and can be analyzed by massive models.

With Vertex AI, developers can build more engaging apps and user experiences by enhancing clustering, anomaly detection, sentiment analysis, and more. These APIs also enable powerful semantic search and text classification functionality, Q&A chatbots based on an organization’s data, and more.

Trusted testers can use the text and picture embedding APIs, which are both available in the preview.

Human Feedback with RLHF

With the help of this innovative tuning tool, businesses may use human input to train a reward model that can be employed to improve foundation models. It is especially helpful in businesses where accuracy is essential, like healthcare or customer contentment is essential, like finance and e-commerce. Ultimately, it results in improved customer satisfaction and engagement.

Additionally, it enables humans to more thoroughly examine model replies for prejudice, harmful information, or other factors, training the model to produce suitable results.

Final Thoughts

With the support of enterprise-grade security, and privacy, these developments represent the next stage in our mission to support developers as they use generative AI technology to innovate bravely and ethically.

Vertex AI users can access Codey, Imagen, Embeddings API for images, and RLHF, while everyone with a Google Cloud account can access Chirp, PaLM 2, Embeddings API, and Generative AI Studio for text in preview. Google believes in constantly transforming how businesses of all sizes and technical proficiency levels create and work with AI in the cloud.