Using Generative AI on GCP: 2023 I/O Extended Mbarara

This session will introduce the audience to generative AI and how it enables us to efficiently make faster decisions, thereby saving cost and time.

It will also introduce the audience to Vertex AI which is the GCP service for generative AI. The session will also explain the end-to-end process of ML-Ops solutions on Google Cloud. The session will end with sample use cases and companies that have leveraged Vertex AI features in serving their customers.

Transcript

1. GDG Mbarara 2023 Using Generative AI on GCP with Vertex AI.
2. GOOGLE IO Extended - SPEAKER TEMPLATE Table of Contents 04 06 12 14 17 19 20 21 24 26 Introduction What is AI? Gen AI How Gen AI Works Transformers Application of Gen AI Building Gen AI on GCP Benefits & Use cases Quick Demo Recommendation
3. About this Session o Introductory guide on gen AI o Good for all levels o Familiarity with computers and the internet required Expected Outcomes o Know what gen AI is and its benefits o Understand how gen AI relates to the AM/ML ecosystem o Know the GCP services available for building gen AI o Understand how to build with Vertex AI
4. Background Information on AI/ML Introduction Section 01
5. A kind of AI technology that can produce a variety of content. o Text o Images o Audio It could also produce synthetic data What is Generative AI Page 5
6. Page 6 A field of CS that deals with making systems that could learn, reason and act autonomously. What is AI?
7. Page 7 A subfield of AI involves training models to recognize patterns in input data and then make predictions or helpful predictions from new data (previously unseen) drawn from the data used to train the model. o Takes input data o Draws new (never seen before) data from the input data o Makes decisions based on the new data What is ML?
8. Requires labelled input data for the training to happen. Humans would labelled the data and pass them to the model for training. Yes labels, yes tags Yes input variables No output variables Examples: Classification i.e identifying data Takes in raw unlabelled input data. It then finds the patterns in the dataset or group the data into groups based on similarities. No labels, no tags Yes input variables Yes output variables Examples: clustering and grouping similar data points Supervised Unsupervised Kinds of ML Models Page 8
9. Page 9 o A subset of ML based on neural networks o It trains the model to process data like the human brain What is DL?
10. Discriminates between different kinds of data instances Trained on a labelled dataset Used for classification or prediction Conditional probability, P(A|B) Example: the model takes in a dataset of cat and dog images and it’s able to classify dogs and cats Generates new data instances Generates new data similar to the one it was trained on Used to generate data Joint probability, P(A,B) Example: the model takes in a dataset of cat images and it is able to generate a new cat image. Discriminative Generative Kinds of DL Models Page 10
11. Generative AI is a subset of Deep Learning NOTE
12. Understanding generative AI and how to build in the Cloud Gen AI Section 02
13. 13 Generative AI Page 13 Don’t lose track: AI > ML > DL > Gen AI o A subset of Deep Learning o Learns patterns from input data to develop new and unique output o Even realistic content at the human intelligence level o Can create new content and ideas Photos, music, video
14. Learns patterns from input data to develop new and unique output Creates a statistical model by learning from the input data Learns pattern Creates model It uses the statistical model to predict expected response to prompts and generates the appropriate content Generate content Gen AI: How it works Page 14
15. ✓ Images ✓ Audio ✓ Text/Speech ❌ Number ❌ Class ❌ Probability Page 15 Emphasis on Content Generation
16. Page 17 PaLM: Pathways Language Model LaMBDA: Language Model for Dialog Applications Noteworthy mentions
17. Gen AI is powered by Transformers. NOTE
18. Transformers ushered in a massive revolution in the NLP ecosystem. A transformer consists of an encoder which encodes the input and a decoder that decodes the representation for the appropriate task. Transformers 2017 Introduced in NLP Page 19 Output I’m alright. Thank you! Encoder Decoder Transformer Model Generative Pre-trained Transformer Model input Hello, how are you?
19. o Code generation o Voice assistants o Personalized learning o CAD Design in Manufacturing o And more Application of Gen AI Page 20
20. o Generative AI Studio: robust platform for building o Generative AI App Builder: no-code platform o Duet AI o Palm API/Makersuite o Vertex AI: model garden of foundation models trained on large amounts of data Building Gen AI on GCP Page 21
21. How you can leverage the technology Benefits and Use cases of Gen AI Section 03
22. o Improved efficiency o Personalized and Customized solutions for businesses & individuals o Better customer experience o Reduced costs o Enhanced creativity Benefits of Gen AI Page 21
23. o Payroll and HR management systems o Using Gen AI to provide powerful business insights for better leadership decision-making Page 23
24. o Drive-thru food ordering fast food chain o Building automated food ordering experience with Google Cloud Gen AI Page 24
25. A Demo with Generative AI Studio in Vertex AI Page 25 console.cloud.google.com/vertex-ai/generative
26. Conclusion Section 04
27. o Discover Generative AI: https://ai.google/discover/generativeai/ o Generative AI learning path: https://www.cloudskillsboost.google/journeys/118 Recommendation Page 26
28. Thank You Jekayin-Oluwa Olabemiwo Software Engineer