Targeting its cloud customers, the e-commerce giant Amazon Web Services unit has proposed to join the hyped generative AI race. AWS declared the launch of two of its own large language models on Thursday.
While one will generate text, the other will facilitate power web search personalization and an array of other crucial tasks.
Amazon, however, says that the company has no plans to release any chatbot like Microsoft’s ChatGPT or Google’s Bard. Amazon has always been focused on machine learning technologies and artificial intelligence. The launch of the said generative AI model, Bedrock and Amazon Titan, will seemingly take this decade-old passion of Amazon to the next level.
AWS says Bedrock will introduce a way to develop an avalanche of AI-powered apps via pre-trained models from startups like StabilityAI, Anthropic, AI21 Labs, etc. Bedrock will now be available in a limited preview. Besides, it will offer access to AWS’s house-trained Titan FMs.
Bedrock’s Debut – The Prelude
Bedrock’s debut is not unprecedented – a few months back, AWS announced its inked partnerships with different generative AI startups. In addition, the company’s annual financial report mentioned its growing investments in tech.
Moreover, StablityAI chose AWS as its preferred cloud provider last November. In March 2023, AWS collaborated with Hugging Face to bring StabilityAI text-generating models onto the AWS platform. Furthermore, AWS collaborated with Nvidia this March to create a next-generation AI infrastructure.
According to AWS, the said infrastructure was designed to reduce the adaption time of machine learning models, which will eventually result in a reduced training cost.
Both AWS and Nvidia have been vocal about the launch of the tools. According to the organizations, those tools will help process user requests in a specific field of code generation, speech recognition, video and images, knowledge, and other generative AI-powered applications.
The Custom Models
The Bedrock will allow AWS customers to use an API to explore AI models from different providers.
The organization declared that Bedrock is designed to cater to an extensive range of enterprise-scale AI apps. This attribute will differentiate Bedrock from its rival AI model hosting services like Azure, Google Cloud, or Replicate. Bedrock features several third-party models hosted on it. They include the Jurassic-2 family from AI21 Labs, Claude from Anthropic, and Stable Diffusion from Stability AI.
The Jurassic-2 family is multilingual and can produce text in Dutch, Italian, Portuguese, French, Spanish, and German. Claude, on the other hand, is capable of performing several text processing and conversational tasks, while Stable Defusion can make art, graphic, images, logos, etc.
AWS is going to provide its customers with the leverage of personalizing any Bedrock model by pointing the service at a few labeled examples in Amazon’s cloud storage plans. The term ‘few’ refers to something as few as 20. Furthermore, Amazon claims they won’t use their customer’s data to train their underlying models.
With this declaration, Amazon is probably trying to induce a sense of security in users, assuring them that there will be the least chance of data security-related complications.
Share
Krishi Chowdhary
Journalist
