Amazon announced the launch of its generative AI (artificial intelligence) assistant Amazon Q targeted specifically at businesses on November 29, the first day of Amazon Web Services’ (AWS) five-day conference re: Invent. Amazon Q relies on a company’s information repositories, code bases, and enterprise systems to answer user queries. Speaking about the AI assistant, AWS CEO Adam Selipsky said that Amazon Q understands roles and permissions within an organization: “If a user does not have permission to access something without Q, they cannot access it with Q either.” Selipsky also pointed out that data acquired by Q from various enterprises will not be used to train AI models. Amazon Q has been trained on 17 years’ worth of AWS knowledge and can provide users with assistance on how to use the AWS management console and other AWS capabilities.
AWS mentioned that Amazon Q is currently available for customers in preview and also available to use through Amazon Connect, the company’s omnichannel cloud contact center service. As a part of Amazon Connect, Q provides customer service agents with recommended responses and actions based on real-time customer questions.
Other major AI announcements made so far in the conference:
Additions to Amazon Bedrock: Selipsky spoke about how the company had launched Amazon Bedrock in September. Bedrock is a fully managed service that makes foundation models (FMs) from multiple AI companies (like Anthorpic, Stability AI, and Meta) available through a single application programming interface (API). During the conference, Selipsky announced that customers would now have options to try out different models and even combine them to meet the needs of their businesses. Users will now have support for fine-tuning Cohere Command, Meta Llama 2, and Amazon Titan models to their businesses, with Anthropic’s Claude also having fine-tuning support soon.
Businesses can also evaluate models in Amazon Bedrock. Companies have two types of evaluation modes at their disposal— human evaluation and automatic evaluation. Automatic evaluation works on predefined metrics such as accuracy, robustness, and toxicity. For more subjective criteria, such as friendliness and alignment to the brand voice, AWS suggests that human evaluation be relied upon.
Guardrails for AI: The company has also announced Guardrails for Amazon Bedrock which allows enterprise customers to implement safeguards and responsible AI policies that are customized to their business. Guardrails help users “define denied topics and content filters to remove undesirable and harmful content from interactions between users and your applications.” This, AWS believes, provides an additional layer of control to companies on top of the protections already enforced by the foundation model developers. Guardrails for Amazon Bedrock are currently only available in preview.
Amazon Titan image generator: Announced during the conference, this image generator creates “realistic images or enhance[s] existing images using natural language prompts for rapid ideation and iteration on large volumes of images and at low cost. ” The company claims that the Amazon Titan Image generator can understand complex prompts and generate relevant images with accurate object composition and limited distortions. Images generated by the tool come with an invisible watermark that the company says is designed to resist alterations.
Amazon Multimodal embeddings: This is a foundational AI model that helps enterprises build more accurate and contextually relevant multimodal search and recommendation experiences for end users. The model has the ability to act on text prompts, image prompts, or a combination of the two. The model converts these prompts into numerical representations to understand semantic meanings and relationships among data.
STAY ON TOP OF TECH NEWS: Our daily newsletter with the top story of the day from MediaNama, delivered to your inbox before 9 AM. Click here to sign up today!