Foundation models are very large machine learning models pretrained on large amounts of data. These models, from AI companies such as AI21 Labs, Anthropic, Cohere, Meta, and Stable Diffusion, are offered via a single API. Additionally, AWS has adding Amazon Titan Embeddings and Meta 2 Llama models to Bedrock.
Use cases for foundation models range from search to content creation to drug discovery, AWS said. An Amazon CodeWhisperer capability, due soon, will provide AI-powered code suggestions. Specific benefits AWS cited for Amazon Bedrock include:
- Choice of leading foundation models.
- Model customisation with a userâ€™s data.
- Native support for RAG (Retrieval-Augmented Generation) to extend the power of foundation models with proprietary data.
- Data security and compliance certifications.
Amazon Bedrock lets customers experiment with a variety of foundation models and customise them privately with proprietary data, AWS said. The cloud provider noted that the service includes managed agents that execute complex business tasks, ranging from booking travel and processing insurance claims to managing inventory, with no need to write code.