AWS re:Invent is positioned to be one of those events of the year that gives a sneak peek into the upcoming technological trends. Gen AI was the most discussed topic during AWS re:Invent 2023.
The industry veterans talked at length about the revolution and transformation that Gen AI will be bringing soon. The informative sessions on the applications of Generative AI in different industry sectors were popular amongst the attendees.
In this blog post, we will highlight some of the key points about Gen AI and AWS solutions, its futuristic applications, groundbreaking innovations and how it will drive the tech landscape. Get prepared to embark on a journey to the future of Gen AI, as the AWS re:Invent sets the stage for it.
Key Takeaways Regarding Gen AI from AWS re:Invent
1. Availability & addition of new foundation models
Amazon Bedrock is a fully managed service that helps in building Generative AI applications – offers you the chance to choose from leading foundation models, and provides sophisticated mechanisms to customize the models with proprietary data & enterprise-grade security.
New Models that are available:
- Anthropic’s Claude 2.1
- Meta Llama 2 70B
- Amazon Titan
- Titan text express
- Titan text lite
- Titan Image Generator (currently in preview)
The Titan Image Generator model is an interesting one. It can help rapidly generate images at low cost, understand complex prompts and generate relevant images with accurate object composition and limited distortions. This can be leveraged to enhance CX and hyper-personalization. Since security & privacy are big concerns AWS Titan will add an invisible watermark to all the generated images to help reduce the spread of disinformation by providing a discreet mechanism to identify AI-generated images.
2. Evaluating foundation models
Enterprises looking to build sophisticated Generative AI-powered applications leveraging some of the general purpose or vertical-specific models that are available, will need to consider which foundation model is a best fit for their requirements based on the use cases, industry they are in or the compliance requirements that they may need to fulfil. To help with this AWS has launched the Model Evaluation feature on Amazon Bedrock.
It is currently in preview, the feature is aimed to simplify several tasks such as identifying benchmarks, running assessments and comparing models for a given task such as question-answer or text summarization. It can also help benchmark models against predefined evaluation criteria such as accuracy, robustness and toxicity tested against proprietary data sets or publicly available data sets.
3. Training & deploying models efficiently at lower cost
Since large foundation models require a lot of computing power, the training and deployment costs are expensive. To address this aspect AWS has introduced two new offerings – SageMaker HyperPod & SageMaker Inference.
- SageMaker HyperPod removes the heavy lifting involved in building and optimizing machine learning infrastructure for training models and reducing training time by up to 40%.
- SageMaker Inference is targeted at helping enterprises reduce model deployment costs and decrease latency in model responses. It deploys multiple models to the same cloud instance to better utilize the underlying accelerators.
4. Amazon Q – Generative AI Assistant
We are aware of how Microsoft with its Copilot and Salesforce with EinsteinGPT have positioned themselves and plugged in Generative AI assistants to all their products. Amazon Q is Amazon’s answer to Microsoft’s Copilot. Amazon Q can be used by enterprises across a variety of functions developing applications, transforming code, generating business intelligence, acting as a generative AI assistant for business applications, and helping customer service agents via the Amazon Connect offering.
- Amazon Q helps developers in building, deploying, and operating workloads on AWS
- Amazon Q helps contact center agents to formulate customer responses and respond to queries quickly and accurately
- Amazon Q helps employees hold conversations, solve problems, generate content and gain insights
5. AWS & Nvidia partnering for Generative AI infrastructure
AWS & Nvidia are expanding their alliance. It includes plans to add supercomputing capabilities to AWS’s AI infrastructure. The biggest initiative is Project Ceiba, a supercomputer that will be hosted by AWS for Nvidia’s R&D.
New Amazon EC2 instances featuring Nvidia GPUs are also in the work. They will offer much more power efficiency. These new instances are aimed at startups, enterprises, and researchers looking to experiment with AI.
There are also plans to integrate NeMo Retriever microservice into AWS to help users with the development of generative AI-powered chatbots. NeMo Retriever is a microservice that enables enterprises to connect custom LLMs and enterprise data.
Mastek's strategy for Generative AI has its core in plan.ai, create.ai, orchestrate.ai, transform.ai, and deliver.ai. This comprehensive approach encompasses various solutions that can be efficiently delivered by leveraging the services outlined above within the AWS technology stack. Whether the goal is to enhance stakeholder experiences, reduce knowledge latency through generative AI-powered chatbots, or handle tasks such as preparing foundational model data, fine-tuning models, and deploying them at scale with reduced costs, these objectives can be accomplished using AWS's Generative AI services. We will continue to monitor developments in this space and align our AWS capability and competence-building strategy accordingly.
GenAI is a rapidly developing field. AWS's Generative AI services can be used to achieve a variety of goals, including improving stakeholder experiences, lowering knowledge latency through chatbots powered by generative AI, and managing tasks like gathering foundational model data, optimizing models, and deploying them at scale at a lower cost. AWS has made significant investments in developing its AI platforms and services. Customers who use the AWS technology stack stand to gain from the use of GenAI.