Re:Invent AWS CEO Adam Selipsky previewed Amazon Q to 50,000 re:Invent attendees in Las Vegas on Tuesday, touting it as a chat-based generative AI assistant whose scope includes suggesting cloud infrastructure to suit business needs, outputting blog posts, helping with application code, and searching and analyzing enterprise data.

Selipsky seemed determined to show that AWS was racing ahead with generative AI equally as fast as any of his competitors - notably, perhaps, Microsoft, whose close relationship with ChatGPT-maker OpenAI is well-known. The Windows giant has now injected chat-based assistants and generative AI, powered by OpenAI's models, into every corner of its empire, from Azure and Bing to its desktop operating systems. By generative AI, we mean models that take natural language prompts or conversations with users, and turn those requests into text, pictures, actions, and so on.

On stage with Selipsky at re:Invent was Anthropic’s CEO and co-founder Dario Amodei. Amazon invested “up to $4 billion” in the AI startup in September, providing Anthropic more cloud infrastructure and chips to train and run its models. Anthropic’s Claude LLM will power a bunch of AWS offerings, including AppFabric.

We asked Amazon if Q will be driven by Claude: we were told the cloud giant isn't revealing the underlying tech. Feel free to speculate. Amodei was up on stage as a customer, apparently.

“The founders of Anthropic were a set of people who worked at OpenAI for several years,” Amodei told the audience. “Seven of us left and founded Anthropic.” Amodei said Claude was ideal for things including content generation, conversation and Q&A. This appearance is fab timing for the upstart, so soon after OpenAI's management drama.

Claude 2.1, the latest build, has doubled the size of its “context window” to 200,000 tokens, which Amodei explained as being equivalent to “a relatively long book.” He added that “a second feature is that we reduce the rate of so-called hallucinations, when models say things that aren’t true, by about two-thirds ... no one is perfect at solving this problem but we are leading the way,” he assured us.

Fast forward an hour or so in what was a lengthy keynote, and Selipsky further expressed his enthusiasm for generative AI. “We believe that generative AI can help everyone at work seamlessly, with helpful, relevant suggestions, whether or not you know the first thing about foundation models,” he said, adding that it has the potential to “transform virtually every customer experience.”

This was part of the lead-in to describing Amazon Q, “a new type of generative AI powered assistant designed to help you at work.” Q, he stated, will use natural language interaction to chat, generate content, and take action. It will be informed by private corporate data, but will respect permissions, he promised. “If the user does not have permission to access something without Q, they cannot access it with Q.”

What will Amazon Q do? It has deep knowledge of AWS services, Selipsky said, so developers can ask questions like “how do I build a web application on AWS, what are my options?” Q will discover more about what is required and use AWS's well-architected patterns to form recommendations. It can also recommend EC2 (Elastic Compute Cloud) instant types for specific workloads, he promised.

Another use case, Selipsky said, is troubleshooting. “If you have an error in the console, you press the troubleshoot with Amazon Q button. Q is going to research the error and suggest how to fix it. Amazon Q also knows about networking, “to help resolve connectivity issues quickly.”

Amazon Q will also surface in CodeWhisperer, a coding assistant which is receiving a lot of attention, with features that sound similar to those offered by Microsoft's programming Copilot, such as explaining code and generating tests.

Another key feature is that Amazon Q will apparently help with migrating source to new language versions, a big issue for organizations with applications stuck on old codebases. Migration is rarely straightforward, but using older code causes performance and security implications.

“We’ve been using this internally,” said Selipsky. “A very small team of Amazon developers successfully upgraded 1,000 applications from Java 8 to Java 17 in just two days.”

He added that Amazon Q will be able to migrate applications from Windows .NET Framework to cross-platform .NET on Linux, a great idea although often challenging in practice because of Windows-only dependencies.

Meet the new expert

Amazon Q in another guise will be “your business expert,” said Selipsky. “Q has connections to over 40 popular enterprise systems, so employees across the organization can ask complex questions and get detailed, accurate, nuanced answers that are relevant to their role.”

Amazon VP of AI Matt Wood explained how Amazon Q can gather data from applications such as “Salesforce, Microsoft 365, Google, Slack and more, supporting over 40 popular business applications.” It will index the content and understand “the concepts, product names, organization structure, all the details that make up your business,” he promised.

A web application will then allow employees to ask questions about that data, such as what are the biggest customer support issues currently. Q can also take and perform actions such as automatically creating tickets in the tracking tool Atlassian Jira.

It goes on: Amazon Q will also integrate with AWS QuickSight, a business intelligence tool, to draw pretty graphs based on organization data.

The assistant will also work with Amazon Connect, a cloud contact center. In fact, when the customer is on the line complaining, Q will be “on the call,” Selipsky said, informing the agent of the details of the case.

“Today, contact center agents spend a lot of time gathering information and customers to understand their questions,” he added, claiming that Amazon Q will speed this task.

Amazon Q is in preview and more details are available here, for general business use; here for IT professionals and developers; QuickSight here; and Connect here. Reading these documents it is obvious that the connectors are a hard problem to solve and likely to be a source of tricky issues for early adopters.

There is also the matter of Q and other LLM systems generating wrong or bad answers to consider, and despite Amodei's acknowledgement that "nobody is perfect at solving this problem," we heard little more about it - although perhaps hallucinations during customer service calls are not exclusive to generative AI.

There are also some pricing details: $20 per user per month for Business Expert, $25 per user/month to add developer and IT professional features, and additional plans for Connect and Quicksight.

An SDK for developers to extend and customize Amazon Q is here and includes API calls such as CreateWebExperience.

Why Q? It stands for Question, we understand, as well as being a reference to Ian Fleming’s Q who made "stealthy, helpful tools" for James Bond. And nothing to do with OpenAI conspiracy theories. ®

https://www.theregister.com//2023/11/28/amazon_q_aws/