Large Language Models, Generative AI & Google Cloud Vertex AI
Google launched Vertex AI 18 May 2021 at Google I/O and it seems like the product has faired well considering all the developments in LLMs and Generative AI
Predictive AI & Generative AI
Large Language Models have two main use-cases, or sides if you like. The one is Generative, and the other side is Predictive.
There has been much hype and use-case development around the Generative side of LLMs; often referred to as Generative AI.
I believe one of the reasons Generative AI has been implemented so widely is due to ease of implementation. Prompt Engineering is the primary vehicle for Generative AI in general and also language in specific.
Google Cloud started at the opposite spectrum of LLMs, by focussing first on predictive prior to generative.
The challenge with the predictive side is you need to train a model based on very specific training data. Predictive systems need a high level of accuracy, especially if the model is implemented for a very specific domain.
What I find interesting is that Google Cloud has only recently launched semi-publically available generative systems. For instance Bard and the Google Generative AI App Builder for creating LLM Apps and Generative Apps (Gen Apps).
Google Research has also done stellar work with the university of Washington on crating a LLM based Prompt Chainer.
Why Predictive Is Important
The Conversational AI and Chatbot fraternity knows the term “intents” all too well. However, in essence intents are merely classifications…and the principle of classification is an integral part of any predictive AI systems.
Intents are just another term for classes and the process of classification.
The basic premise of classifications is that you have pre-defined classes which are named and these classes each have a set of training data. This allows you to train a model and predict the classification of a user utterance. This could be a single utterance, or a longer piece of text.
More On Vertex AI
Below are the various classification tasks in Vertex AI.
The classification tasks are segmented between images, tabular data, text video.
When selecting the TEXT tab, four options are available for classification. Single label and multi-label text classification, entity extraction and sentiment analysis/classification…
Generative AI
Below is an extract from Google, and how they see the future of Generative AI
Google Cloud will launch a range of products that infuse generative AI into our offerings, empowering developers to responsibly build with enterprise-level safety, security, and privacy. This journey starts today with the introduction of two new technologies:
Generative AI support in Vertex AI gives data science teams access to foundation models from Google and others, letting them build and customise atop these models on the same platform they use for homegrown ML models and MLOps.
Generative AI App Builder allows developers to quickly ship new experiences including bots, chat interfaces, custom search engines, digital assistants, and more. Developers have API access to Google’s foundation models and can use out-of-the-box templates to jumpstart the creation of gen apps in minutes or hours.
In Closing
In upcoming articles I will share prototypes and practical lessons learnt from building predictive models with Google Vertex AI.
⭐️ Please follow me on LinkedIn for updates on Conversational AI ⭐️
I’m currently the Chief Evangelist @ HumanFirst. I explore and write about all things at the intersection of AI and language; ranging from LLMs, Chatbots, Voicebots, Development Frameworks, Data-Centric latent spaces and more.
