AWS re:Invent 2023 - Keynote with Dr. Swami Sivasubramanian

1. Generative AI is augmenting human productivity and fueling intelligence and creativity.

🥈85 01:17

Generative AI is creating a powerful relationship between humans and technology, where both parties contribute to new innovations.

• Generative AI is similar to symbiotic relationships observed in nature.
• Humans and AI together can form new possibilities and co-evolve.

2. Ada Lovelace recognized the potential of computers beyond number crunching.

🥇92 03:55

Ada Lovelace believed that computers could read symbols and perform logical operations, leading to complex tasks.

• She speculated that computers could understand musical notation and create music.
• However, she emphasized that true creativity and intelligence originate from humans.

3. The symbiotic relationship between data, generative AI, and humans drives progress.

🥈88 06:50

The explosion of data enables foundational models to exist, powering generative AI and accelerating innovations.

• Data, generative AI, and humans together unleash creativity and innovation.
• AWS provides a comprehensive AI and ML stack to support this relationship.

4. Amazon Bedrock offers a broad set of capabilities for building and scaling generative AI applications.

🥈82 10:16

Bedrock provides access to leading foundational models, allowing customers to choose models that meet their unique needs.

• Customers can select from a wide range of models from providers like AI21Labs and Stability.
• Bedrock also offers tools for multimodal embeddings and text generation use cases.

5. Titan Multimodal Embeddings enable richer multimodal search and recommendation options.

🥈89 17:34

Titan Multimodal Embeddings allow developers to combine image and text to create more accurate and contextually relevant search experiences.

• Companies like OfferUp and Shutterstock are using Titan Multimodal Embeddings to enhance their search capabilities.
• Amazon also offers models for text generation and image generation use cases.

6. Titan Image Generator allows the production of high-quality realistic images.

🥇91 19:59

Titan Image Generator enables customers to generate or enhance images using natural language prompts and their own data.

• The model is trained on diverse datasets and includes built-in mitigations for toxicity and bias.
• Titan Image Generator promotes responsible AI usage and customization.

7. AWS has developed invisible watermarks for AI-generated images.

🥈85 20:48

AWS has integrated invisible watermarks into its AI-generated images to help reduce the spread of misinformation and provide a discreet mechanism for identification.

• The watermarks are tamper-resistant and designed to be integrated into image outputs.
• This technology is one of the first widely released built-in invisible watermarks.

8. The image generator and model editing features of AWS AI technology.

🥉78 21:20

AWS AI technology allows users to generate images based on text prompts and easily swap out backgrounds.

• Users can use the image generator to create lifestyle images or variations of the original subject.
• The model also offers features like inpainting and image customizations.

9. Intuit's use of AWS and AI to power its financial technology platform.

🥇92 25:06

Intuit has leveraged AWS and AI to build an AI-driven expert platform that powers financial technology applications.

• Intuit uses AWS services like Sagemaker and Bedrock to scale its data capabilities and machine learning platform.
• They have built a proprietary GenAI operating system called Genomics to enable personalized and accurate experiences for their customers.

10. The importance of data in customizing AI models.

🥈86 34:51

Customizing AI models with labeled datasets and fine-tuning allows businesses to create personalized and accurate experiences.

• Data is the key to unlocking accuracy and building unique customer experiences.
• Fine-tuning and leveraging unlabeled datasets can help businesses adapt models to their specific needs.

11. The benefits of fine-tuning and using third-party models on Bedrock.

🥈82 36:23

Fine-tuning models and using third-party models on Bedrock provide businesses with the flexibility and optionality to build the best customer experiences.

• Fine-tuning allows businesses to customize models with their own data and domain-specific knowledge.
• Using third-party models on Bedrock ensures access to best-in-class solutions for specific tasks.

12. Retrieval Augmented Generation (RAG) augments prompts with contextual information.

🥈85 37:29

RAG is a technique that enhances prompts with contextual information from private data sources, resulting in more accurate and relevant responses.

• Developers need to convert their data into vector embeddings and store them in a vector database.
• Knowledge Bases for Amazon Bedrock simplifies the process of ingesting, retrieving, and augmenting prompts with contextual information.

13. Knowledge Bases for Amazon Bedrock supports the entire RAG workflow.

🥈82 38:37

Knowledge Bases for Amazon Bedrock simplifies the process of ingesting, retrieving, and augmenting prompts with contextual information.

• Developers can point to the location of their data, and Bedrock fetches relevant context and text documents.
• It supports popular vector databases like Vector Engine for OpenSearch, Serverless Redis, Enterprise Cloud, and Pinecone.

14. Agents for Amazon Bedrock enable GenAI applications to execute complex tasks.

🥈88 40:03

Agents for Amazon Bedrock allow developers to dynamically invoke APIs and connect to internal systems and APIs on behalf of users.

• Developers can define instructions and orchestrate models to access data sources.
• Agents simplify the process of fulfilling user requests and executing business tasks.

15. GenAI capabilities can be leveraged for various tasks, such as DIY projects.

🥈86 40:40

GenAI-powered assistants can provide accurate and easy-to-follow steps for DIY projects.

• Users can ask natural language questions and receive detailed steps, materials, and tools.
• GenAI-powered assistants can generate images and summaries of user reviews to aid decision-making.

16. SageMaker Hyper Pods reduce model training time and improve efficiency.

🥇91 48:58

SageMaker Hyper Pods enable distributed training across thousands of chips, reducing model training time by up to 40%.

• Hyper Pods automatically take checkpoints, resume training after hardware failures, and optimize model performance.
• They provide a significant increase in training throughput and handle distributed capacity efficiently.

17. Perplexity leverages AWS services for conversational answer engine.

🥈89 51:26

Perplexity uses AWS services like Bedrock, SageMaker, and Hyper Pods to build and deploy conversational answer engines.

• Perplexity fine-tunes open-source models and orchestrates multiple models in one product.
• They benefit from AWS customized services for training and inference, improving accuracy and efficiency.

18. A strong data foundation for GenAI applications includes comprehensive and integrated services.

🥈85 56:14

A strong data foundation for GenAI applications requires access to a comprehensive set of services that can handle the scale, volume, and type of data. AWS offers a broad range of tools for storing, organizing, and accessing various types of data.

• AWS provides a wide selection of database services, including relational and non-relational databases.
• AWS also offers tools for machine learning, analytics, and data warehousing.
• Vector capabilities have been added to popular data sources like Amazon Aurora, Amazon RDS, and OpenSearch Service.

19. Zero ETL integrations make it easier to access and analyze data across different sources.

🥇92 58:14

AWS has invested in building seamless integrations across its data stores, enabling customers to break down data silos and create a more integrated data foundation. Zero ETL integrations allow for near real-time analytics and eliminate the need for complex ETL pipelines.

• Zero ETL integrations have been added to services like Aurora, Redshift, DynamoDB, and OpenSearch Service.
• The integration between Amazon OpenSearch and S3 enables seamless search, analysis, and visualization of log data.
• Customers can securely share data with partners using Clean Rooms, enabling collaborative analysis without sharing the whole dataset.

20. Data governance and security are essential for a strong data foundation.

🥈88 1:09:36

To ensure a high-quality and compliant data foundation, data needs to be secured and governed throughout the development of applications. AWS offers services like Amazon DataZone for cataloging and governing data, as well as Clean Rooms for secure data sharing.

• Amazon DataZone helps organizations catalog, discover, share, and govern data.
• Clean Rooms enable secure data sharing with partners without sharing the underlying data.
• Clean Rooms ML allows for the application of ML models with partners without sharing the underlying data.

21. Booking.com is a two-sided marketplace for travel accommodations.

🥈85 1:13:40

Booking.com is not just a platform for booking accommodations, flights, rental cars, and attractions, but also a two-sided marketplace with partners all over the globe.

• Booking.com manages over 28 million listings of places to stay.
• They offer flights in 54 countries and rental cars in over 52,000 locations worldwide.
• They also provide attractions booking services.

22. Booking.com partners with AWS to tackle data challenges.

🥇92 1:15:07

Booking.com recognized the need for help in managing their massive amount of data and partnered with AWS to address the challenges.

• They manage over 150PB of data.
• With AWS, their data scientists can train more jobs concurrently, decrease job failures, and reduce training time.
• They also leverage AWS technology to build AI trip planner and personalized hotel recommendation options.

23. Booking.com uses generative AI for conversational trip planning.

🥈88 1:15:53

Booking.com has built an AI trip planner that allows users to book a trip in a conversational manner.

• They use the Llama 2 model for intent detection.
• The AI trip planner moderates conversations and protects customer privacy.
• They leverage their review data and recommendation engine to provide personalized hotel recommendations.

24. AWS leverages AI to optimize data management and analytics.

🥇91 1:19:14

AWS uses AI to optimize data management and analytics, making it easier to use and more intuitive.

• They optimize data warehouse performance with AI-driven scaling and optimizations.
• They use AI to simplify data querying and provide customized SQL recommendations.
• They also leverage AI for data integration and building data pipelines.

25. AWS enables zero ETL integrations for data management.

🥈89 1:20:53

AWS offers zero ETL integrations to simplify data management tasks and eliminate the need for custom ETL jobs.

• They provide tools like Redshift Query Editor and Amazon MQ for data querying and integration.
• They leverage AI to optimize data warehouse performance and automate data integration tasks.
• They also offer natural language interfaces for data integration and troubleshooting.

26. AWS enables data-driven innovation with AI and ML.

🥈87 1:27:36

AWS combines data management and AI/ML technologies to spur net new innovation.

• They use AI and ML to transform data management and analytics.
• They leverage AI to optimize data warehouse performance and automate data integration.
• They enable data-driven decision-making and storytelling with zero ETL integrations.

27. Using Amazon Q in QuickSight to measure success in SAAS.

🥈85 1:31:11

Amazon Q in QuickSight is used to measure key metrics and critical data for success in SAAS businesses.

• It provides an executive summary of important insights.
• It helps identify the impact of new features on customer experience.

28. Creating customizable data stories with Amazon Q.

🥈88 1:31:50

Amazon Q allows users to create customizable data stories based on actual business data.

• Users can select the format, visuals, and build the story in seconds.
• The stories cover the problem and impact to customers.

29. Huron AI's mission to improve cancer care access.

🥇92 1:33:52

Huron AI aims to make cancer care accessible to everyone, regardless of location.

• They have created applications to improve cancer care in countries with limited oncologists.
• Their innovations fill critical cancer data gaps for underrepresented populations.

30. Toyota's use of data and AI for safety and customer experiences.

🥈89 1:38:02

Toyota utilizes data and AI to improve vehicle safety and customer experiences.

• They collect data from sensors in vehicles to determine if a collision has occurred.
• They use generative AI to develop an AI-powered assistant for vehicle owners.

31. Model evaluation and selection with Amazon Bedrock.

🥈86 1:45:06

Amazon Bedrock offers model evaluation and selection capabilities to optimize AI models.

• Users can evaluate and compare models based on qualitative and quantitative criteria.
• Human review workflows and comprehensive reports are provided.

32. Building an app with Party Rock is easy and fun.

🥈85 1:48:40

You can build your own app with Party Rock in a few steps, experiment with different prompt engineering techniques, and add widgets like a chatbot to make your application more useful and fun.

• You can select different models to see what works best for your use case.
• Once you are happy with the results, you can publish your application and invite others to use or remix it.

33. The power of data and human creativity in innovation.

🥇92 1:50:12

The symbiotic relationship between data, GenAI, and humans is accelerating our ability to create new innovations and differentiated experiences.

• Each individual brings unique inputs and ideas to the table.
• The combination of data and human creativity is key to unlocking transformative technology.