Amazon Web Services (AWS) added to its AI holdings today with new Amazon Bedrock offerings that are positioned to assist companies with developing generative AI applications and experiences.
With large organizations like the New York Stock Exchange and Ryanair relying heavily on Amazon Bedrock services, it s clear that this platform holds a lot of value. However, many of the updates announced today by Amazon are meant to help organizations of any size create and improve generative AI applications in a more granular and security-focused manner.
Custom Model Import and Model Evaluation
One of the biggest changes announced to the Amazon Bedrock platform is Custom Model Import, which allows organizations to import and access their own custom models as fully managed Application Programming Interfaces (APIs) within bedrock. While Bedrock already offers well-known models from companies like Anthropic, Meta, and Mistral AI, certain customers have demands that reach outside what these companies can provide.
Specifically, Amazon mentions that customers in healthcare, financial services, and other industries with heavy security and regulatory needs will benefit from Custom Model Import. Many organizations within these industries are beginning to put their own data to work by customizing publicly available data with the help of Amazon SageMaker. These models can then be easily added to Amazon Bedrock.
With Custom Model Import, these same companies can import and access their own custom models as APIs in Bedrock. These customers can take models that they customized on SageMaker, or even other tools, and add them to the Bedrock platform. This allows companies to choose a combination of Amazon Bedrock models and their own custom models via the same API. As of right now, Custom Model Import on Amazon Bedrock is available in preview and supports three of the most popular open model architectures; Flan-T5, Llama, and Mistral.
On top of this, AWS is announcing that Model Evaluation is now generally available on Amazon Bedrock. By selecting predefined evaluation criteria – such as accuracy and robustness – and then uploading their prompt library or selecting publicly available datasets, customers can quickly decide which premade model is right for them.
Guardrails
As AI tools explode into every aspect of modern commerce, people are wondering how to take advantage of this technology in a safe and responsible way. This is where Guardrails for Bedrock comes in.
Now generally available, Guardrails is the only solution offered by a major cloud provider that allows customers to have built-in and custom safeguards in a single offering. It also works with all large language models in Amazon Bedrock, as well as fine-tuned models. What’s more, AWS states that Guardrails helps customers block up to 85% of harmful content.
What’s more, creating Guardrails is quite simple. Customers must provide a natural-language description outlining the defined topics within the context of their applications. Customers are also given the option to configure thresholds to filter important areas like hate speech, sexual language, insults, prompt injection, and violence.
Amazon s new Amazon Bedrock offerings provide a comprehensive platform for companies to develop secure and customizable generative AI applications, addressing the growing demand for AI-powered solutions across various industries.
#AI/ML/DL #Cloud #Software #Amazon #AmazonBedrock #AmazonSageMaker #AWS [Source: EnterpriseAI]