Generative AI models like ChatGPT have taken the tech world by storm because of their potential to penetrate the mainstream. With the release of Bedrock, it's clear that Amazon is ready to go all-in, just like its big tech rivals from Microsoft and Google.
Bedrock will allow Amazon Web Services (AWS) users to build generative AI from base models (FMs). GPT-4 is an example of such a model, and ChatGPT is a generative AI application built on top of it. According to a blog post announcing the service, Bedrock is a "serverless experience" where users can "privately customize FMs with their own data, and easily integrate and deploy them into their applications."
To coincide with the Bedrock launch, Amazon also released Titan, which includes two new base models developed by Amazon Machine Learning. Details about the Titan are scarce at this time, with Amazon representatives keeping the technical specifications under wraps. However, AWS vice president Bratin Saha told reporters that Amazon has been using a "fine-tuned version" of Titan to display search results on the company's home page.
Users won’t be limited to Amazon’s internal FM, though, as the company also announced Bedrock integrations for some of the industry’s most popular models, including Jurassic-2, a multilingual language learning model (LLM), and Claude, Anthropic’s conversational agent built on top of the company’s The "Constitutional Artificial Intelligence" basis.
Bedrock will also provide platform API access to Stability AI models, including Stable Diffusion, a popular text-to-image generator. Amazon may be a little late, with GPT-4 still a month away, but with AWS being nearly ubiquitous and easy to use, Bedrock and Titan could give the incumbent industry leaders a run for their money - using what it offers.
The cost of training a generative AI model can be staggering. Unfortunately, once a model is trained on a given dataset, it is essentially "dirty" data and may tend to "hallucinate" information from it in response to irrelevant queries. Bedrock allows users to solve this problem by giving them the option to use a pre-existing FM as the backbone backing their data. For customers already on AWS, and those bringing data into the AWS ecosystem, this means their data remains as secure as it normally is on the Amazon cloud, and is never injected into the training dataset.
According to Amazon's announcement, "None of the customer's data will be used to train the underlying models, and since all data is encrypted and never leaves the customer's virtual private cloud (VPC), customers can trust that their data will remain private and Confidential."
Amazon hopes to turn the generative AI showdown between Google’s Bard and Microsoft/OpenAI’s ChatGPT into a battle royale by announcing its Bedrock service and the debut of two new in-house large LLMs. Another Chinese tech company, Baidu, tried to break through with the launch of Ernie a tribute to Google Bud but the product’s poor reception sent the company’s shares tumbling 10%. As Titan and Bedrock prepare to join the first wave of generative AI models available to the public, tech companies around the world are preparing to jump in. Chinese tech company Alibaba is launching its own AI chatbot called Uniform Qianwen, hoping to bring some competition to the Western companies that currently dominate the field.
