AWS Announces AI and Generative Offerings to Compete with Microsoft

Table of Contents

• AWS Unveils LLM and Image Generation Solutions
• Retrieval Augmented Generation Gets Boost
• Introducing Amazon Q for Business Use
• Multimodal Search Coming Soon
• Other AWS Re:Invent Highlights

AWS Unveils LLM and Image Generation Solutions

At its annual Re:Invent conference, Amazon Web Services (AWS) announced several new AI offerings aimed at competing directly with Microsoft. These include new large language models (LLMs), image generators, and an AI assistant called Amazon Q.

The flagship announcement was the launch of AWS's proprietary LLM called Titan. Like other foundation models such as GPT-3 and PaLM, Titan is designed to understand and generate natural language. AWS also unveiled Titan Image Generator, which can create images from text prompts but with a unique built-in digital watermarking capability.

Proprietary Titan LLM and Image Generator

The Titan LLM and Titan Image Generator are AWS's first proprietary foundation models. The image generator includes both visible and invisible statistical watermarks. AWS claims these allow the content's provenance and authenticity to be tracked. In addition to its own models, AWS announced partnerships with AI safety companies like Anthropic to integrate their products with AWS offerings. Constitutional AI's Claude was highlighted as an example.

Partnerships with Anthropic and Others

AWS emphasized its collaborations with AI safety startups. Besides Anthropic, companies like Hugging Face and Cohere were mentioned as partners. By partnering, AWS hopes to gain credibility in responsible AI practices while startups can reach AWS's existing large enterprise customer base.

Retrieval Augmented Generation Gets Boost

AWS introduced new capabilities aimed at improving retrieval augmented generation (RAG). RAG systems connect an LLM to external knowledge sources like databases.

Specifically, AWS is simplifying the process of indexing databases and encoding them as vectors to make data retrieval through semantic search faster and more streamlined.

The goal is to break down data silos so LLMs have access to more contextual information, improving their ability to hold nuanced conversations and generate high-quality content.

Introducing Amazon Q for Business Use

The biggest announcement was Amazon Q, a new AI assistant service tailored for business applications.

Amazon Q integrates natural language conversation and generation abilities with data from popular business systems. It allows companies to stand up customized AI tools aligned with branding guidelines and access controls.

Potential use cases highlighted include transforming press releases into blog posts, analyzing campaign results, and generating social media prompts based on business needs.

Multimodal Search Coming Soon

AWS previewed upcoming capabilities around multimodal search, allowing queries combining text, images, audio or video.

A furniture retailer could enable shoppers to search for sofas using a photo for example. AWS even demonstrated a party planning use case, with Alexa not only providing advice but suggesting relevant products available on Amazon.

Multimodal search lays the foundation for tighter integration between AI assistants and ecommerce or recommendation engines.

Other AWS Re:Invent Highlights

AWS made several other AI announcements at Re:Invent:

• Stability AI unveiled an image generation model called SDXL Turbo focused on near real-time performance crucial for interactive applications

• DeepMind open sourced data on 2.2 million theoretical materials discovered by its graph neural network tool GNoME, which could accelerate new material science discoveries

FAQ

Q: What is Titan from AWS?
A: Titan is AWS's new proprietary large language model that can generate text. It also has an associated image generator that produces images with built-in watermarks.

Q: What is retrieval augmented generation?
A: Retrieval augmented generation (RAG) involves using vector databases to retrieve knowledge and data to inform and improve text generation from large language models.

Q: What is Amazon Q?
A: Amazon Q is a new AI-powered generative assistant service tailored for business use cases like content creation, analysis, and recommendations.

Q: What is multimodal search?
A: Multimodal search allows users to search using multiple modes like text, images, audio, and video to find relevant results.

Q: How does Amazon Q align responses?
A: Amazon Q can align responses from its large language model with company brand standards and messaging.

Q: What can you use Amazon Q for?
A: Example business applications of Amazon Q include transforming press releases into blog posts, creating social media prompts, and analyzing campaign results.

Q: How was Amazon involved with the woodshed example?
A: The hypothetical example suggested Amazon could use multimodal search and connections with Amazon shopping to recommend and deliver all necessary supplies to build a woodshed based on budget and other criteria provided.

Q: What is SDXL Turbo?
A: SDXL Turbo is a new text-to-image generation model from Stability AI that enables real-time image generation for improved speed and interactivity.

Q: What materials were discovered by GNoME?
A: The GNoME tool from DeepMind simulated 2.2 million materials, with 380,000 theoretically stable, that could have applications in batteries, circuit boards, and other areas.

Q: How were the materials discovered?
A: The materials were discovered using a graph neural network system trained on molecular structures and chemistry.