In the past, only large technology companies, government agencies, and universities had the ability, time, and money to deploy generative AI capabilities. In the past two decades, cloud services have lowered the threshold for computing. Dr. Matt Wood, Vice President of Global Products at Amazon Cloud Technology, said: "We hope to do the same with generative AI and provide technology to every developer and business user."
A strong base model is crucial. The Amazon Titan proposed by Amazon Cloud Technology is a collection of a series of large models, including self-developed models and many industry-leading third-party large models.
Make AI development faster and lower the threshold
In addition to the large model, Amazon Cloud Technology's existing set of systems also plays a role in the practice of refactoring all products.
After selecting a model and writing code on Amazon Bedrock, people can use high-performance infrastructure to train and run their own models, including Amazon EC2 Inf1 instances supported by Amazon Cloud Technology Inferentia and Amazon EC2 Trn1 instances supported by Amazon Cloud Technology Trainium And Amazon EC2 P5 instances powered by Nvidia H100 Tensor Core GPUs.
At the data level, Amazon Cloud Technology has built an end-to-end cloud-native data strategy, allowing people to obtain data insights more conveniently and safely. Take advantage of Amazon DataZone, released last year. People can catalog, discover, share and govern data stored in the cloud, on-premises and third-party sources more quickly and easily, while providing more granular control tools to manage and govern data access rights and ensure data security.
In addition, customers can also use Amazon SageMaker to build, train and deploy their own models, or use Amazon SageMaker Jumpstart to deploy popular basic models, including Cohere's large language model, TII's Falcon 40B and Hugging Face's BLOOM.
On top of the basic model, framework, and computing power, there are tools that directly help developers. Amazon CodeWhisperer is an artificial intelligence-driven code generation service. After training on billions of lines of publicly available open source code and Amazon Cloud Technology's own code base, it can automatically generate Java, JavaScript, Python, and other and other codes in 15 languages.
When using Amazon CodeWhisperer, AI extracts useful resources from other source codes to generate new code, which makes it conform to the habits of specific developers. After the AI generates the code, it can automatically check the software license, so that the code can be directly used.
Amazon Cloud Technology said that compared with similar products such as Copilot, Amazon CodeWhisperer can provide effective and safe codes, and more importantly, it does not have any usage restrictions, and individual users are completely free.
AI writing code is a fascinating concept for programmers, but it has been unclear whether it can actually improve productivity. Amazon Cloud Technology conducted experiments on two groups of engineers internally, and the results proved that using Amazon CodeWhisperer can increase the work speed by 57%, and the quality of the code can also be improved by 27%.
If you don't use code at all, you can also use large models to build applications. At the scene, Amazon Cloud Technology also demonstrated the ability to build financial management applications using only natural language prompts without using code at all.
In the future, there will be more opportunities to work side by side with AI, and people's standards for programming will change. Maybe a programmer who can't use Amazon CodeWhisperer is not a good programmer?
After ChatGPT, generative AI technology is still developing rapidly. For infrastructure providers, Amazon Cloud Technology is also open to the future. Matt Wood said that Amazon Titan will introduce more industry-leading open source and Third-party large models allow developers to use the most advanced capabilities.
On the other hand, there are some things that are certain: Amazon Cloud Technology hopes to help developers, startups, and technology giants break down the threshold of using large models by continuously achieving high availability of AI infrastructure.
"I believe that Amazon Bedrock will continue to lower its price so that users can use it out of the box and pay on demand, eliminating the huge cost of purchasing computing hardware. In the future, users of large models can pay according to the number of input words and output words , Whether it’s a startup or a technology giant, we’ll make Amazon Bedrock’s cost-effective service accessible to everyone,” said Matt Wood.
Since the release of ChatGPT at the end of last year, global technology giants have competed fiercely for technologies such as large language models, and solutions for the implementation of large model technologies have emerged in an endless stream. Amazon Cloud Technology, which released Amazon Titan and Amazon Bedrock services in April this year, is considered a latecomer. However, in terms of building an AI platform, only Amazon Cloud Technology can provide a complete set of software and hardware infrastructure, attract many top startups, integrate advanced technologies and form services.
In the marathon of generative AI, the goal of Amazon cloud technology is to get the first place on the finish line. With the development of technology and people's continuous application practice, it is foreseeable that the technology of Amazon cloud technology will also give birth to more AI applications.