The cloud computing giant, Amazon Web Services (AWS), was seen as lagging behind competitors like Microsoft Azure and Google Cloud in the emerging field of generative AI until recently. However, during the past two days at its AWS Re:Invent conference, Amazon has aggressively positioned itself to lead in supporting enterprise companies building generative AI projects, particularly challenging Microsoft. Here are the key announcements made today:
- Expanded LLM Choice: Through its Bedrock service, AWS provides access to various pretrained foundation models, including its own Titan model, third-party models like AI21’s Jurassic and Anthropic’s Claude. AWS announced support for Anthropic’s Claude 2.1, boasting a 200K token context window and improved accuracy.
- Multi-Modal Vector Embeddings: AWS introduced Titan Multi-model Embeddings to facilitate multimodal search and recommendation within LLMs.
- Titan TextLite and Titan TextExpress: These text generation models, designed for tasks like summarization and open-ended text generation, are now generally available.
- Titan Image Generator (Preview): This model creates realistic images with invisible watermarks for security and customization, with a focus on mitigating toxicity and bias.
- Simplified Retrieval-Augmented Generation (RAG): AWS unveiled KnowledgeBase for Amazon Bedrock, enabling easier integration of proprietary data for LLMs. This includes support for popular databases like Amazon Aurora and MongoDB.
- Model Evaluation on Amazon Bedrock (Preview): Companies can now evaluate and compare foundation models for their specific use cases.
- RAG DIY “Agent” App: AWS showcased RAG DIY, an AI assistant that helps users with various projects, leveraging natural language questions, image generation, and more.
- Gen AI Innovation Center: AWS will provide custom support for building models, particularly around Anthropic’s Claude models, to help enterprises fine-tune models using their own data.
- Sagemaker Hyperpod for Model Training (GA): AWS introduced Hyperpod to simplify and accelerate the training of foundation models.
- Database Integration and Vector Support: AWS is working to break down silos between its databases, including vector data support, to improve accessibility for LLMs.
- Vector Search for Redis (Preview): AWS introduced vector search support for in-memory DB for Redis, catering to security-focused use cases like fraud detection and real-time chatbots.
- Neptune Analytics (GA): Combining vector search with graph analytics in Amazon Neptune allows for deeper insights and relationships discovery in interconnected data.
- Machine Learning on Cleanroom Data (Preview): AWS introduced AWS Clean Rooms ML, enabling data sharing with third parties for predictive insights through machine learning.
- Amazon Q for Generative SQL in Amazon Redshift (Preview): Amazon Q now supports SQL, allowing users to use natural language prompts for customized SQL queries in Redshift lakehouse. It will soon support data integration pipelines using natural language with Amazon Glue.
These announcements demonstrate Amazon’s commitment to becoming a dominant player in the generative AI space and providing diverse solutions to enterprise customers.