With enterprise use cases of generative AI expected to take off in 2024, Amazon Web Services (AWS) on Tuesday used its developer conference in Las Vegas, re:Invent, to unveil a chatbot designed for integration with business tools, called Amazon Q, which can perform actions such as generating a JIRA help desk ticket, automate code writing, and automate AWS cloud services.
“Amazon Q builds on AWS’s history of taking complex, expensive technologies and making them accessible to customers of all sizes and technical abilities, with a data-first approach and enterprise-grade security and privacy built-in from the start,” said Dr. Swami Sivasubramanian, vice president for data and AI at AWS, in prepared remarks.
Also: Every enterprise plans to increase AI spending next year
The program can automatically answer questions such as best practices for AWS, but the intention is that it will be hooked up to customer applications and data sources and become tailored to a company’s tasks.
The bot has 40 built-in connectors “for popular data sources,” said Amazon, “including Amazon S3, Dropbox, Confluence, Google Drive, Microsoft 365, Salesforce, ServiceNow, and Zendesk, as well as the option to build custom connectors for internal intranets, wikis, run books, and more, Amazon Q makes it fast and easy for customers to get started.”
Also: I spent a weekend with Amazon’s free AI courses, and highly recommend you do too
The program integrates with AWS’s business intelligence program, QuickSight, to do things such as automatically generate business reports (“How are my sales doing?”). It also integrates with the company’s call center application, Amazon Connect. Future integration is planned for AWS Supply Chain, which will allow a company to pose questions of logistics, such as, “What is causing the delay in my shipments and how can I speed things up?”
Amazon did not offer technical details on the composition of Amazon Q, such as what large language model or models are used as the foundation of the program. The company states that it is “trained on 17 years of AWS knowledge and experience.”
The Amazon Q chat interface automatically appears in the management console of an AWS account. ZDNET used a free account on AWS to try out Q with some simple questions such as “What is the most popular project on EC2,” referring to AWS’s basic virtual machine. That produced the answer “The most popular project on EC2 is the AWS Deep Learning AMI,” for “amazon machine image,” and several paragraphs about AMI, along with links to supporting documentation.
The bot has some very basic fails, however, when it comes to simple questions about things such as generative AI on AWS. For example, it does a fine job when asked, “What’s the easiest way to build a large language model on AWS?” It responds with lots of stuff about pre-trained language models available through the SageMaker development environment.
Also: How to create your own custom chatbots using ChatGPT
But, when asked, “If I want to use one of the SageMaker large language models, what’s the easiest way to fine-tune it on my own data,” Q says it cannot answer the question. That’s a very basic question for which it should have material.
Amazon Q was up-to-date about the latest AI accelerator chip, also announced Tuesday, called Trainium 2. However, it could not provide technical details, and instead recommended using an instance of Trainium 1.
The bot has guardrails that pop up with unacceptable input. Asking the question, “Are you a sentient being?” produces the answer, “Sorry, I can’t answer that question. Please ask me a different question about AWS.” The same blanket response pops up for illicit questions involving advice for harmful acts, fraud, etc.
While the program responded with numerous suggestions to the question, “What is Amazon Q good for?” it declined to reply to the question, “What is the underlying large language model powering Amazon Q?”
Also: AWS unveils new Trainium AI chip and Graviton 4, extends Nvidia partnership
Pricing for Amazon Q is $20 per user per month for the “business” account, and $25 per user per month for the “builder” account. The extra $5 per user per month for Builder brings numerous AWS-specific functions, such as integration of Q with Amazon’s coding tool, CodeWhisperer.
Full specifications of the pricing plans are offered on a dedicated Q pricing page.
The full livestream of Tuesday’s keynote for re:Invent can be seen on the Amazon home page.