– AWS launches Custom Model Import feature in Bedrock, allowing organizations to import and access their custom generative AI models as fully managed APIs
– AWS offers tools like Guardrails and Model Evaluation to help customers fine-tune and evaluate their models for things like hate speech or privacy issues
– Titan Image Generator and Titan Text Embeddings V2 are upgraded and released, providing better creativity for image generation and more efficient text-to-numerical representation modeling, respectively
AWS, Amazon’s cloud computing business, is focusing on becoming the preferred platform for companies to host and enhance their custom generative AI models. They recently introduced Custom Model Import as part of Bedrock, their suite of generative AI services, enabling organizations to bring in their proprietary models and manage them as APIs within AWS infrastructure. This feature allows companies to fine-tune their models, expand their knowledge, and mitigate any biases present in the models.
With the majority of enterprises building and refining their own generative AI models, AWS aims to address the need for cloud compute infrastructure for these companies through Custom Model Import. The feature offers model customization options such as Guardrails for filtering model outputs and Model Evaluation for testing the model’s performance. These options, along with the ability to experiment with multiple models using the same workflows, make Custom Model Import an attractive offering for AWS customers.
While services like Vertex AI and Microsoft’s Azure AI development tools offer similar model evaluation features, AWS’s Bedrock distinguishes itself with its Titan family of generative AI models. The newly upgraded Titan Image Generator, for example, can create images from text descriptions or customize existing images with enhanced creativity. Additionally, the Titan Text Embeddings V2 model has improved efficiency and accuracy compared to its predecessor, making it a cost-effective option for text-to-numerical representation applications.
As AWS continues to develop its generative AI capabilities, they aim to address ethical concerns around deepfakes by incorporating tamper-resistant watermarks in images created with their Titan models. The company also emphasizes its commitment to customer indemnification in the event of copyright infringement related to AI-generated content. Despite a lack of transparency regarding training data sources, AWS assures customers of their efforts to ensure ethical and legal compliance in their generative AI offerings.