Author | Zheng Siyu
Generative AI and Big Language Modeling (LLM) are undoubtedly the biggest hot spots in the AI industry today. Since the beginning of 2023, generative AI technologies such as ChatGPT and Stable Diffusion have attracted a lot of attention and application boom in the industry and at the public level with their amazing output performance.
On June 27-28, 2023, the Amazon Web Services China Summit was held at the Shanghai World Expo Center. On the first day of the conference, Dr. Matt Wood, Vice President of Global Products of Amazon Web Services, delivered a keynote speech to convey Amazon Web Services' attitude and vision to generative AI technology. Matt Wood said that generative AI is a huge breakthrough for the entire technology industry, and Amazon Web Services is excited about it. In order to allow more people to benefit from this technological innovation earlier, Amazon Web Services is helping enterprises, developers, and even the entire industry to use the technology, resources, and experience accumulated by Amazon Web Services to promote the popularization of generative AI technology.
Generative AI: How far is the industry from the future?
The foundation of generative AI is large-scale models with hundreds of billions of parameters, trained on large datasets of text, code, or images, requiring tens of thousands of GPUs or more, and training in thousands of hours. Big models are able to unlock innovations that transcend traditional boundaries and push humanity into uncharted territory. Users can use large models to accelerate the creative process, speed up the summary sequencing process, or create new interactive experiences, and even improve the reliability and efficiency of complex decisions.
However, while generative AI technology has broad application prospects, it also faces obstacles on the way to adoption. The huge hardware resources, time, and labor cost required for large model training are heavy burdens for most enterprises. Enterprises need to collect massive amounts of data when training models, and how to deal with the legal authorization of these data and how to ensure the security of data will also cause headaches for many enterprise development teams. At this stage, the output quality of large models is still a problem, and public services including ChatGPT have shown problems such as unstable output quality, overconfident answers, and insufficient understanding of domain knowledge in practice.
In response to the above problems, Amazon Web Services also proposed corresponding solutions at this summit. In his speech, Dr. Matt Wood mentioned several key products and services that Amazon Web Services is involved in generative AI: Amazon Bedrock, Amazon Titan, and Amazon CodeWhisperer. These products, combined with Amazon Web Services' existing Amazon SageMaker, Amazon Aurora and other services, as well as Amazon Web Services' self-developed Inferentia and Trainium chips, will provide the industry with a path to the popularization of generative AI for civilians.
Large model is democratic, the ideal and solution of Amazon Web Services
Amazon Bedrock is a fully managed generative AI service just launched by Amazon Web Services that allows users to access pre-trained base models from Amazon Web Services and third-party base model providers through APIs. Developers don't need to worry about the underlying infrastructure, just select the desired model through an easy-to-use API and output the appropriate text or image content. Amazon Web Services also provides the Amazon Titian Model Library, which allows users to optimize and fine-tune models in a secure, private environment. Finally, the Amazon CodeWhisperer tool uses generative AI technology to help users significantly improve development efficiency, quickly reap the benefits of generative AI, and lower the threshold for large model applications.
Matt Wood mentioned that since the beginning of cloud computing, Amazon Web Services has been driving the democratization of innovative technology. In the field of generative AI, Amazon Web Services believes that the key to democratizing large models is to have a low threshold for use, rich variety, and low development and application costs, easy to customize models for industry and domain needs, and at the same time be safe and reliable, without worrying about legal and privacy issues. To this end, Amazon Bedrock and Amazon Titan provide a set of foundational models with their own expertise, allowing users to easily enter data into the model and deploy applications through serverless APIs. Amazon Bedrock supports private data customization, providing developers with a secure development environment.
For enterprises, Amazon Bedrock has created a very suitable generative AI starting point and development framework for them, and they can quickly adjust the models in the Amazon Titan model library to industry models that can better solve domain problems, greatly reducing the development threshold of large models. Amazon Bedrock's pay-as-you-go fee model, combined with Amazon Web Services' self-developed high-performance inference and training chips, can greatly reduce the cost of large-scale model training in the early stage. Matt Wood believes that this approach is simple and affordable enough to meet the needs of both large and small and medium-sized venture capital companies.
It is worth mentioning that Amazon Bedrock can also greatly reduce the probability of large model output full of confidence but wrong answers, and the output content has been carefully reviewed in the cloud to ensure health and harmlessness. When Amazon Web Services trains Amazon Titan models, the data used is also authorized or licensed, which meets relevant legal requirements, and enterprises can use these models with confidence without worrying about potential privacy and legal issues.
Today, companies have used Amazon Bedrock to develop industry models for ad content distribution and have achieved 50% cost savings, 35% efficiency gains, and 45% CTR improvements. For the foreseeable future, a large number of industry enterprises will realize the benefits of the generative AI services solutions offered by Amazon Web Services and enjoy similar benefits through these solutions.
Unleash the potential of AI: from front-office practice to back-office support
While developing the foundational framework services for generative AI, Amazon Web Services is also focusing on how to improve productivity with generative AI, such as the new Amazon CodeWhisperer. It can provide great help to developers, and users can instruct the system to generate the code they need through natural language, greatly improving development efficiency. Amazon CodeWhisperer supports 15 programming languages, with more options to be added in the future. The tool is able to extract useful resources from the source code libraries used by developers to generate code to more accurately grasp requirements.
In internal comparisons, CodeWhisperer has resulted in 57% development time savings and 27% code quality improvements, and the effect is so clear that enterprises, including Swiss Army Knife, have started using Amazon CodeWhisperer to improve their existing software development processes. Amazon Web Services believes that Amazon CodeWhisperer can also help users further lower the threshold for large-model technology applications, allowing them to develop generative AI applications in a faster, higher quality, and more secure way, helping the process of technology democratization.
Matt Wood mentioned that the essence of generative AI is to support and process data in ways that have never been done before. Data is the starting point for everything for generative AI, and Amazon Web Services' answer to this is a cloud-native data strategy. If Amazon CodeWhisperer is the front-end application for generative AI, then cloud-native data strategy is the back-end support that Amazon Web Services provides for enterprises.
First, Amazon Web Services provides the world's leading all-category cloud database service, providing low latency and low cost. For example, metaverse enterprise Gevos uses Amazon Web Services' cloud database as a hosting platform for core workloads, significantly improving the productivity of game development. At the same time, users have access to a broad and deep portfolio of tools such as Amazon EMR, Amazon Aurora's new ETL service, and more. Finally, data management services like Amazon DataZone can help organizations build a data governance framework that can shape their own data strategy.
All of these data capabilities combine to give Amazon Web Services customers a visible first-mover advantage. When users take full advantage of these capabilities to combine with AI services such as Amazon SageMaker and Amazon Bedrock, they can quickly implement new applications and extract considerable value from existing data.
As users drive data and AI strategies, Amazon Web Services also sets up enough security barriers for enterprises to help development teams gain more freedom and conduct more extensive experiments while avoiding security and legal issues. The results of these experiments allow developers to combine expertise in their industries to deliver satisfying industry AI applications.
Write at the end
This Amazon Web Services China Summit marks that Amazon Web Services has officially begun to make a comprehensive attack in the field of generative AI, and has made great strides towards the goal of full popularization of this technology. Dr. Matt Wood's presentation not only comprehensively reviewed a series of innovative products and services of Amazon Web Services, but also provided good suggestions for enterprises to build their own generative AI development frameworks and develop AI and data strategies using these services.
Matt Wood concludes that across the industry, Amazon Web Services enables the fastest, lowest cost, and easiest way to deliver generative AI models. Amazon Web Services has also recently invested $100 million to further promote innovation and advancements in generative AI, including the establishment of a new Generative AI Innovation Center. This innovation hub brings together Amazon Web Services machine learning scientists while actively working with customers to help them envision, design, and launch new generative AI products, services, and processes.
Matt Wood believes that as more businesses and even individuals regularly use these generative AI models, their feedback and output will get better. Dr. Matt Wood is also optimistic about the future of the industry: "We are in the early stages of a major technological innovation and on the eve of the explosion, and everything we see is just the starting line, not the ceiling. ”