Amazon Web Services (AWS) introduced a new feature called Guardrails API during its annual summit in New York. This API, previously only available on Amazon Bedrock, is now standalone and allows users to set up guardrails for AI applications, regardless of whether they are using Amazon Bedrock or not. Vasi Philomin, AWS’s vice president of generative AI, highlighted the importance of guardrails in ensuring security, privacy, and veracity in AI models. The goal of the Guardrails API is to offer customers more model options and flexibility, even supporting OpenAI models.
Bedrock, AWS’s cloud service, released in April 2023, hosts various AI models but does not currently support OpenAI models. Railings, a feature within Bedrock, offers similar functionalities to other model gardens like Microsoft’s Azure AI, catering to enterprise customers who need more flexibility in interpreting certain terms.
The Guardrails API is in preview and includes a feature called Contextual Grounding, which helps detect hallucinations in models or AI systems by ensuring that responses are grounded in real data. This feature can filter out more than 75% of hallucinations, providing a layer of protection for users.
AWS also announced enhancements for building AI agents on Bedrock, including increased memory retention for agents across interactions and a new code interpretation capability for analyzing complex data with code. These updates aim to provide users with more customization options and choices when working with AI models, aligning with Amazon’s vision of empowering customers with the tools to leverage generative AI in various industries.
Overall, AWS is focused on giving customers the capabilities needed to overcome challenges and implement AI solutions effectively. Through features like the Guardrails API and improvements to Bedrock, AWS is paving the way for a more robust and adaptable AI ecosystem for its users.
Article Source
https://venturebeat.com/ai/aws-makes-guardrails-a-standalone-api-as-it-updates-bedrock/