Bedrock, a new AWS cloud service, allows developers to build and scale generative AI chatbots and other applications in the cloud, using internal organizational data to fine-tune on a variety of leading pre-trained large language models (LLMs) from Anthropic, AI21, Stability AI, as well as two new LLMs in Amazon’s Titan model family.
Amazon CEO Andy Jassy spoke directly to the AWS focus on enterprise AI with Bedrock when speaking to CNBC’s “Squawk Box” yesterday.
“Most companies want to use these large language models but the really good ones take billions of dollars to train and many years and most companies don’t want to go through that,” he said. “So what they want to do is they want to work off of a foundational model that’s big and great already and then have the ability to customize it for their own purposes. And that’s what Bedrock is.”
According to Gartner analyst Sid Nag, with the buzz and excitement around generative AI news from Google and Microsoft, Amazon was overdue to follow suit.
“Amazon had to do something,” he told VentureBeat in an interview. “The cloud providers are obviously best suited to handle data-heavy generative AI, because they are the ones that have these hyper-scale cloud computing storage offerings.”
Bedrock, he explained, provides a meta layer of usability for foundation models on AWS. Amazon is also notably calling out its ability to provide a secure environment for organizations to use this type of AI, he added. “Organizations want to create their own walled garden in a generative AI model, so I think you’ll see more and more of that,” he said.
Bedrock’s multiple models makes Amazon’s AWS attractive
Emad Mostaque, CEO of Stability AI, pointed out that Bedrock’s offering of multiple models including Stable Diffusion plays to Amazon’s history of focusing on choice. “In his original plan to $100 billion of revenue, Jeff Bezos envisioned that half that revenue would be Amazon products and half third party through their marketplace,” he told VentureBeat in a message.
While it may have been surprising that Cohere was not on the list of a Bedrock models — it is available on SageMaker and AWS — CEO Aidan Gomez said the company decided not to participate in the Bedrock product at this time. “We may change our opinion and join the ‘model zoo’ in the future, but we decided not to be a part of this initial release, he told VentureBeat by email.
But Yoav Shoham, co-founder and co-CEO of AI21 Labs, focused on the fact that AWS has curated a set of best-in-class models. “There is a class of text-based applications particularly well served by Jurassic-2’s multilingual, multi-sized models,” he told VentureBeat by email. “We look forward to enabling, jointly with AWS, the creation of many such applications.”
Low-code platform Pega was noted in AWS VP Swami Sivasubramanian’s blog post yesterday as one of Bedrock’s early adoptors. Peter van der Putten, director of the AI Lab at Pega, said the company intends to use Bedrock for a range of use cases in our platform, which they will make available to its customers.
“For example, just based on a simple sentence such as ‘create a dental insurance claim application,’ we can generate a runnable prototype low code app including workflow, data models and other artifacts, which will jumpstart, democratize and accelerate development of low code business applications,” he said. “There are also other areas in our low code platform where we leverage it, such as allowing users to ask for reports just using natural language.”
The desire for multi-cloud will keep the cloud AI competition going
What makes Amazon very attractive for Pega and its customers, he added, is Bedrock’s access to a wide range of models, commercial as well as open source, in “a safe, enterprise scale manner,” he said. But he also called out the importance of multi-cloud options: “In addition to this, our clients will also be able to access OpenAI models through Azure and we are in discussion with other major cloud players as well, plus keeping a close eye on open source, for the most sensitive applications.”
That, says Sid Nag, is the irony of the cloud AI wars.
“The fundamental premise of building a generative AI model is democratization of data — the more information you have, the higher the fidelity of the response,” he said. “But the whole philosophy and approach that cloud providers have historically taken is ‘I should own everything, everything should run in my estate. So on the one hand, they want to be very predatory, but on the other hand, are they willing to share data across multiple estates?”