A revolution in productivity is already brewing. Global management consulting firm McKinsey pointed out in its recent report "The Economic Potential of Generative Artificial Intelligence: The Next Wave of Productivity" that generative AI could add $2.6 trillion to $4.4 trillion in value to the global economy each year. At the Amazon Cloud Technology New York Summit a few days ago, "generative AI" was also the most frequently mentioned keyword in the audience.
"Today, large models can be pretrained on large amounts of unlabeled data out-of-the-box to handle a variety of general problems. Furthermore, they can be used for domain-specific applications.” said Swami Sivasubramanian, Global Vice President of Database, Data Analytics, and Machine Learning at Amazon Cloud Technologies, “The ability to easily customize pre-trained models through fine-tuning is definitely a game-changer.”
A revolution in productivity is already brewing. Global management consulting firm McKinsey pointed out in its recent report "The Economic Potential of Generative Artificial Intelligence: The Next Wave of Productivity" that generative AI could add $2.6 trillion to $4.4 trillion in value to the global economy each year. At the Amazon Cloud Technology New York Summit a few days ago, "generative AI" was also the most frequently mentioned keyword in the audience.
"Today, large models can be pretrained on large amounts of unlabeled data out-of-the-box to handle a variety of general problems. Furthermore, they can be used for domain-specific applications.” said Swami Sivasubramanian, Global Vice President of Database, Data Analytics, and Machine Learning at Amazon Cloud Technologies, “The ability to easily customize pre-trained models through fine-tuning is definitely a game-changer.”
Relying on customer demand insight and technology accumulation in the past few years, Amazon Cloud Technology has integrated a large number of AI capabilities into easy-to-use products, hoping to deliver technological progress to all walks of life in the simplest way. At this technology event, Amazon Cloud Technology launched seven new generative AI functions in one go.
The strongest generative AI large model, easily called here
In April this year, Amazon Cloud Technology released the fully managed basic model service "Amazon Bedrock", joining the battle of large models as a "key infrastructure provider".
From the perspective of enterprises that want to apply large models, self-developed large models require billions of dollars and years of training. A better solution is to customize and fine-tune some already very powerful open source basic models to meet their own needs. Diverse business needs. Therein lies the great value of Amazon Bedrock. This service allows everyone to start building generative AI applications based on existing large models, dedicated AI computing power and tools, combined with their own data.
In the latest expansion of Amazon Bedrock, the latest results from a group of top large-scale model suppliers are brought together:
Currently, Amazon Bedrock provides access to Anthropic's latest language model Claude 2, AI21's JURASSIC-2, and Amazon's self-developed Amazon Titan series models. Stability AI also launched the latest version of the Vincent graph model suite Stable Diffusion XL 1.0 in Amazon Bedrock. In addition, Cohere became the latest basic model supplier to join Amazon Bedrock, and brought the text generation model Command and the text understanding model EMBED.
Compared with other one-stop large model service platforms, the advantage of Amazon Bedrock is that users can integrate it with the rest of the Amazon cloud technology platform to more easily access the data stored in the Amazon S3 object storage service , and be able to benefit from Amazon Cloud Technology access control and governance policies.
Generative AI has changed the competitive landscape of cloud computing services. In addition to the original storage, computing, and network infrastructure, the provision of capabilities at the model, framework, and application levels has become more important. In the past period of time, we have seen the birth of a new business called "model as a service". A series of platforms like Amazon Bedrock are turning large models into directly usable services, helping users from all walks of life to access generative AI, leveraging a brand new blue ocean market.
Promoting generative AI to complete the "last mile" of landing
Today, even though generative AI models are so powerful, they still cannot replace humans to "perform" some key, personalized tasks. This is precisely a very critical step in the process of transforming "generative AI" into "productivity".
The problem is not insoluble: models can often be attached with APIs, plugins, databases to extend functionality and automate certain tasks for users. For example, ChatGPT has launched a plug-in mechanism before, and also provides an open platform for developers, allowing more users to expand according to their needs, ideas and professional capabilities. In order to simplify the work required in this link, Amazon Cloud Technology officially launched Amazon Bedrock Agents.
Amazon Bedrock Agents can extend the underlying model to understand user requests, break down complex tasks into steps, conduct conversations to gather additional information, and take actions to fulfill user requests. Developers can create fully managed Agents with just a few clicks:
The conversational intelligent agent created by this function can provide personalized latest answers and perform actions based on proprietary data, helping enterprises to accelerate the delivery of generative AI applications and promote the solution of the "last mile" problem of generative AI landing. For example, a business could use Amazon Bedrock Agents to create a customer service chatbot that can process orders, using its internal information (including customer profiles and return policies) to customize service for each order.
If you want to use a metaphor, Amazon Bedrock Agents are like a right-hand man. Perhaps in the near future, we will be able to enjoy the user-side services provided by this function: not only displaying suitable flights and recommending restaurants with good reputation, but also directly helping to book and follow up.
Search Technology Revolution in the Generative AI Era
In the hot discussion of solving the challenge of implementing large models, the concepts of "vector search" and "vector database" have become more and more familiar. This is the transformation that is taking place at the retrieval technology level in the era of generative AI.
First of all, with the growth of data scale, keyword retrieval can no longer meet the demand, and vector retrieval can be used as a supplement to traditional search technology. By representing data as vectors, models can quickly analyze and understand large amounts of information, accurately identifying and matching similar items.
Secondly, although the pre-trained large model has outstanding capabilities, it also has some shortcomings, such as lack of domain knowledge, lack of long-term memory, and lack of factual consistency. In the current situation of growing data scale and increasingly precious computing power, vector databases can be used as the "super brain" of large models to write a cheat sheet and supplement dynamic knowledge at a relatively low cost to meet the growing needs of users.
For this direction, Amazon Cloud Technology has made early efforts, and has launched a number of data storage services that support vectors, including Amazon Aurora PostgreSQL-compatible relational databases, and Amazon RDS (Amazon Relational Database Service) relational databases that are compatible with PostgreSQL.
At this summit, Amazon Cloud Technology launched a vector engine for Amazon OpenSearch Serverless. The vector engine supports simple API calls and can be used to store and query billions of Embeddings.
The engine is powered by the k-Nearest Neighbor (kNN) search function in the Amazon OpenSearch project, providing customers with semantic search services in a serverless environment. Even as vectors grow from a few thousand during prototyping to hundreds of millions or more, the engine scales seamlessly without re-indexing or reloading data to scale the infrastructure.
In response to the extensive needs of the large-scale model era, Amazon Cloud Technology also officially announced that all databases on the platform will have vector functions in the future to help customers simplify operations and facilitate data integration.
Multiple accelerations for generative AI
In addition to these major releases, in order to accelerate the training and application of generative AI, Amazon Cloud Technology has launched a series of services and tools.
The latest development is that two key services have been officially available: one of the services is about computing infrastructure, Amazon EC2 P5 instances based on Nvidia H100 Tensor Core GPU have been officially available to meet customers' requirements for high performance and high scalability when running workloads sexual needs.
Many well-known generative AI models in the industry also cover functions such as question answering, code generation, video and image generation, and speech recognition. The scale usually has hundreds of billions or trillions of parameters, and the training time even takes several months. This is bound to become one of the factors that generally affect the speed of generative AI landing.
Compared with the previous generation of GPU-based instances, Amazon EC2 P5 instances can speed up training by up to 6 times. The training time that used to take days can be shortened to hours, helping customers reduce training costs by up to 40%.
Another service is related to development tools. Last year, Amazon Cloud Technology launched the preview version of Amazon CodeWhisperer, an AI programming assistant, which has attracted great attention from developers. Data shows that users complete tasks an average of 57% faster than developers who do not use the programming assistant. Amazon CodeWhisperer is now available and integrated with Amazon Glue.
From then on, developers can write specific tasks in natural language, and Amazon CodeWhispere will directly recommend one or more code snippets that can complete this task in Amazon Glue Notebooks. Users can choose to "accept the most recommended suggestion", "see more Suggestion" or "Continue to write code yourself". In other words, even if you don't know how to write code at all, you can try to build a complete application using the "human speaking" method.
write at the end
The development of technology often exceeds people's assumptions. Once upon a time, researchers needed to spend months on data preparation, data processing and model training, and had to invest extremely high costs just to complete a specific task.
In this wave, the route chosen by Amazon Cloud Technology is to do a good job as a "key infrastructure provider". Its advantage lies in its deep accumulation of artificial intelligence technology in the past 20 years, its deep understanding of more than 100,000 customers, and its high-availability and powerful infrastructure that has been polished for many years. These will strongly promote Amazon cloud technology to accelerate the journey of generative AI landing, helping every developer or entrepreneurial team.
Relying on customer demand insight and technology accumulation in the past few years, Amazon Cloud Technology has integrated a large number of AI capabilities into easy-to-use products, hoping to deliver technological progress to all walks of life in the simplest way. At this technology event, Amazon Cloud Technology launched seven new generative AI functions in one go.
The strongest generative AI large model, easily called here
In April this year, Amazon Cloud Technology released the fully managed basic model service "Amazon Bedrock", joining the battle of large models as a "key infrastructure provider".
From the perspective of enterprises that want to apply large models, self-developed large models require billions of dollars and years of training. A better solution is to customize and fine-tune some already very powerful open source basic models to meet their own needs. Diverse business needs. Therein lies the great value of Amazon Bedrock. This service allows everyone to start building generative AI applications based on existing large models, dedicated AI computing power and tools, combined with their own data.
In the latest expansion of Amazon Bedrock, the latest results from a group of top large-scale model suppliers are brought together:
Currently, Amazon Bedrock provides access to Anthropic's latest language model Claude 2, AI21's JURASSIC-2, and Amazon's self-developed Amazon Titan series models. Stability AI also launched the latest version of the Vincent graph model suite Stable Diffusion XL 1.0 in Amazon Bedrock. In addition, Cohere became the latest basic model supplier to join Amazon Bedrock, and brought the text generation model Command and the text understanding model EMBED.
Compared with other one-stop large model service platforms, the advantage of Amazon Bedrock is that users can integrate it with the rest of the Amazon cloud technology platform to more easily access the data stored in the Amazon S3 object storage service , and be able to benefit from Amazon Cloud Technology access control and governance policies.
Generative AI has changed the competitive landscape of cloud computing services. In addition to the original storage, computing, and network infrastructure, the provision of capabilities at the model, framework, and application levels has become more important. In the past period of time, we have seen the birth of a new business called "model as a service". A series of platforms like Amazon Bedrock are turning large models into directly usable services, helping users from all walks of life to access generative AI, leveraging a brand new blue ocean market.
Promoting generative AI to complete the "last mile" of landing
Today, even though generative AI models are so powerful, they still cannot replace humans to "perform" some key, personalized tasks. This is precisely a very critical step in the process of transforming "generative AI" into "productivity".
The problem is not insoluble: models can often be attached with APIs, plugins, databases to extend functionality and automate certain tasks for users. For example, ChatGPT has launched a plug-in mechanism before, and also provides an open platform for developers, allowing more users to expand according to their needs, ideas and professional capabilities. In order to simplify the work required in this link, Amazon Cloud Technology officially launched Amazon Bedrock Agents.
Amazon Bedrock Agents can extend the underlying model to understand user requests, break down complex tasks into steps, conduct conversations to gather additional information, and take actions to fulfill user requests. Developers can create fully managed Agents with just a few clicks:
The conversational intelligent agent created by this function can provide personalized latest answers and perform actions based on proprietary data, helping enterprises to accelerate the delivery of generative AI applications and promote the solution of the "last mile" problem of generative AI landing. For example, a business could use Amazon Bedrock Agents to create a customer service chatbot that can process orders, using its internal information (including customer profiles and return policies) to customize service for each order.
If you want to use a metaphor, Amazon Bedrock Agents are like a right-hand man. Perhaps in the near future, we will be able to enjoy the user-side services provided by this function: not only displaying suitable flights and recommending restaurants with good reputation, but also directly helping to book and follow up.
Search Technology Revolution in the Generative AI Era
In the hot discussion of solving the challenge of implementing large models, the concepts of "vector search" and "vector database" have become more and more familiar. This is the transformation that is taking place at the retrieval technology level in the era of generative AI.
First of all, with the growth of data scale, keyword retrieval can no longer meet the demand, and vector retrieval can be used as a supplement to traditional search technology. By representing data as vectors, models can quickly analyze and understand large amounts of information, accurately identifying and matching similar items.
Secondly, although the pre-trained large model has outstanding capabilities, it also has some shortcomings, such as lack of domain knowledge, lack of long-term memory, and lack of factual consistency. In the current situation of growing data scale and increasingly precious computing power, vector databases can be used as the "super brain" of large models to write a cheat sheet and supplement dynamic knowledge at a relatively low cost to meet the growing needs of users.
For this direction, Amazon Cloud Technology has made early efforts, and has launched a number of data storage services that support vectors, including Amazon Aurora PostgreSQL-compatible relational databases, and Amazon RDS (Amazon Relational Database Service) relational databases that are compatible with PostgreSQL.
At this summit, Amazon Cloud Technology launched a vector engine for Amazon OpenSearch Serverless. The vector engine supports simple API calls and can be used to store and query billions of Embeddings.
The engine is powered by the k-Nearest Neighbor (kNN) search function in the Amazon OpenSearch project, providing customers with semantic search services in a serverless environment. Even as vectors grow from a few thousand during prototyping to hundreds of millions or more, the engine scales seamlessly without re-indexing or reloading data to scale the infrastructure.
In response to the extensive needs of the large-scale model era, Amazon Cloud Technology also officially announced that all databases on the platform will have vector functions in the future to help customers simplify operations and facilitate data integration.
Multiple accelerations for generative AI
In addition to these major releases, in order to accelerate the training and application of generative AI, Amazon Cloud Technology has launched a series of services and tools.
The latest development is that two key services have been officially available: one of the services is about computing infrastructure, Amazon EC2 P5 instances based on Nvidia H100 Tensor Core GPU have been officially available to meet customers' requirements for high performance and high scalability when running workloads sexual needs.
Many well-known generative AI models in the industry also cover functions such as question answering, code generation, video and image generation, and speech recognition. The scale usually has hundreds of billions or trillions of parameters, and the training time even takes several months. This is bound to become one of the factors that generally affect the speed of generative AI landing.
Compared with the previous generation of GPU-based instances, Amazon EC2 P5 instances can speed up training by up to 6 times. The training time that used to take days can be shortened to hours, helping customers reduce training costs by up to 40%.
Another service is related to development tools. Last year, Amazon Cloud Technology launched the preview version of Amazon CodeWhisperer, an AI programming assistant, which has attracted great attention from developers. Data shows that users complete tasks an average of 57% faster than developers who do not use the programming assistant. Amazon CodeWhisperer is now available and integrated with Amazon Glue.
From then on, developers can write specific tasks in natural language, and Amazon CodeWhispere will directly recommend one or more code snippets that can complete this task in Amazon Glue Notebooks. Users can choose to "accept the most recommended suggestion", "see more Suggestion" or "Continue to write code yourself". In other words, even if you don't know how to write code at all, you can try to build a complete application using the "human speaking" method.
write at the end
The development of technology often exceeds people's assumptions. Once upon a time, researchers needed to spend months on data preparation, data processing and model training, and had to invest extremely high costs just to complete a specific task.
In this wave, the route chosen by Amazon Cloud Technology is to do a good job as a "key infrastructure provider". Its advantage lies in its deep accumulation of artificial intelligence technology in the past 20 years, its deep understanding of more than 100,000 customers, and its high-availability and powerful infrastructure that has been polished for many years. These will strongly promote Amazon cloud technology to accelerate the journey of generative AI landing, helping every developer or entrepreneurial team.