If you’re reading this article, I can assume that you have previously heard of (or at least are interested) in knowing more about Generative AI. And you’re not alone.
According to Google Trends, the search term “Generative AI” has seen a notorious spike in interest since the beginning of the year, which means that it is clearly a very hot topic in today’s world, with impacts across all sectors of society. (Have you seen the news that NVIDIA – the main GPU player in the market – briefly reached a 1 trillion dollar valuation? Source: Reuters )
In this article, I will offer you an explanation of the main buzzwords related to Generative AI and how they relate with each other, an overview of the current Generative AI offers in Google Cloud Platform, and some resources that you can use in your own learning journey.
Hope you enjoy this article and it is helpful in your exploration into the Generative AI world!
Let’s start from the beginning
Currently, in the media, a lot of terms related to AI are thrown around without context or explanation, so let me start at the beginning and provide some definitions of terms you surely have heard before.
Artificial Intelligence (AI) refers to the field of study related to intelligent systems (systems that simulate human knowledge) that don’t need explicit programming to derive knowledge. We should think of AI as we think of Chemistry, Physics or History.
Machine Learning (or ML) refers to a particular branch of AI where intelligent systems derive knowledge from patterns in the underlying data. Other well-known branches of AI are (but not limited to) Optimization, Computer Vision and Robotics.
Deep Learning (or DL) refers to a subset of ML models whose structure is based on Artificial Neural Networks. Other subsets of ML models that are very well known include (but not limited to) tree-based models (such as Random Forests or XGBoost) or clustering methods.
Generative AI (GenAI) refers to a type of artificial neural network that can generate new data based on the training dataset it was modeled after. Other types of ANNs are discriminative NNs, which attribute labels or classify the new data that it is shown based on the previously learned patterns on the training dataset. Inside Generative AI, we can find dedicated models for language (for example LLMs – Large Language Models), audio, video, image, and 3D, amongst many other fields.
Note: Although it refers to the underlying data used to create these models, another concept that is worth mentioning is structured vs unstructured data. Structured data refers to data that is highly organized, usually in a tabular fashion such as the organization found inside relational databases. Unstructured data refers to data that is not so easily cataloged due to its lack of predetermined formats, such as images, video or audio.
How can you use Generative AI models in GCP?
Earlier in May of this year, Google Cloud Platform made available in preview it’s Generative AI Studio inside Vertex AI. Google defines Generative AI Studio as “a Google Cloud console tool for rapidly prototyping and testing generative AI models.” [1] This means that you don’t need to develop your own Generative AI models from scratch (which can be quite a complex task, not to mention time-consuming), and you can start experimenting and prototyping within GCP with Google’s pre-trained models to assess if a given approach is truly right for your use case.
Currently, inside Generative AI Studio you can find two main usage scenarios:
- Language: with prepared models to solve tasks such as Summarization, Classification, Extraction, Writing, and Ideation (which contains my personal favorite, a Meme Generator).
- Speech: with prepared models to solve tasks such as Speech-to-Text and Text-to-Speech.
With these pre-trained models, you can solve several recurrent tasks such as Sentiment Analysis, Contract Analysis, or even create a chat agent that is able to automatically summarise previous interactions and create follow-up lists. You can also fine-tune these pre-trained models to the specific needs of your use cases, without the need of recreating the whole model from scratch.
It is also available in the preview of the new Model Garden, which contains several foundational models that are enterprise-ready and task-specific. You can find first-party models directly created by Google (such as PaLM 2), open-source models (such as BERT) that are trusted in the community and, in the future, Google hopes to make available third-party models as well.
In the near future, something very exciting is coming: Image related scenarios.
We expect to see offerings in image generation, classification, search, and recommendation. Some lucky testers already have a preview of this incredible feature, so hopefully, it will be available soon. As a teaser, I’ll leave you with the image shared by Google to present this tool.
Fonte: Vertex AI Image Generation – Image generated using Vertex AI Image Generation from the prompt: 4K video game concept art, urban jungle, cityscape inspired by new york city, detailed rendering.
Resources
Since Generative AI is becoming widely available and is such a hot topic, Google offers a learning path to help you acquire new skills in this field at no cost.
You can find the Generative AI Learning Path here, and it covers topics such as (but not limited to):
- Large Language Models
- Responsible AI
- Image Generation
- Attention Mechanism
- Transformer Models
This Learning Path is a great way to learn the concepts in depth from Google itself and have direct contact with these tools in the Google Cloud Platform environment by completing the labs.
Conclusion
First of all, thank you for spending the last few minutes of your time reading my overview of Generative AI and Google Cloud Platform’s tools for Generative AI. Hopefully, I left you with a clearer understanding of the buzzwords and a sparked curiosity about this field.
Generative AI is growing at an incredibly fast pace, and sometimes it feels like every day a new product is hitting the market, which in turn means that everyday new and exciting advances are being made. But it might also mean that it is time to reflect on ethics and responsible AI.
On that topic, I will leave you today with Google’s Responsible AI business case, as well as Google’s Responsible AI Practices, which you will hopefully read and implement in your day-to-day use of Generative AI.
