During Amazon’s Q1 earnings call, CEO Andy Jassy announced that the company is currently working on an advanced Large Language Model (LLM) that will improve Alexa’s capabilities beyond simple tasks such as setting alarms and playing music. Similar to ChatGPT, it is a deep learning algorithm that can generate human-like text-based responses and other content based on user instructions.
Amazon’s move may be in response to criticism from third-party device vendors who have claimed that Alexa’s capabilities have remained stagnant. In addition, it is also to develop a more advanced LLM that aligns with their continued commitment to Artificial Intelligence (AI), which has also been reflected in recent earnings calls from other tech giants such as Google, Microsoft, and Meta. Google CEO Sundar Pichai stated that Google would continue to incorporate AI to enhance search capabilities. Meanwhile, Microsoft CEO Satya Nadella announced that the company would invest further in AI, citing an increase in usage for Bing after integrating ChatGPT. Finally, Meta CEO Mark Zuckerberg confirmed that the company plans to invest in AI and will introduce new AI-related updates across its various apps.
In a Q&A session with Brian Nowak, the managing director at Morgan Stanley, Jassy expressed his concern that popular generative AI tools like ChatGPT and Microsoft 365 Copilot are overshadowing Alexa’s position as the primary personal assistant. His comments highlight the intense competition in the virtual assistant market and the need for Alexa to remain competitive by constantly improving its capabilities.
He also acknowledged that while the company’s current LLM has been powering Alexa, they are now working on developing an even more capable LLM. Jassy believes that this enhanced LLM will help Amazon achieve its goal of creating the best personal assistant in the world.
However, Jassy also considers the fact that achieving this goal across multiple domains will be challenging. Building an LLM that can excel in a wide range of areas, from music and entertainment to productivity and communication, requires significant investment and expertise
Jassy adds that Amazon is already in a strong position with Alexa, which has hundreds of millions of endpoints being used across a wide range of domains, from entertainment and shopping to smart homes and information. He further explains Amazon has had an LLM underneath it, “but we’re building one that’s much larger and much more generalized and capable. And I think that’s going to really rapidly accelerate our vision of becoming the world’s best personal assistant. I think there’s a significant business model underneath it.”
Jassy emphasized that Amazon has been heavily investing in LLMs for several years, as well as in customized machine learning chips, particularly GPUs, that are optimized for LLM workloads. For instance, Trainium chips are optimized for machine learning training, and Inferentia chips are specialized for inference or the predictions that come from the model. He also pointed out that the company recently released second versions of Trainium and Inferentia.
According to Jassy, the combination of price and performance that these chips offer is significant and differentiated from competitors. He believes that a lot of machine learning training and inference will run on AWS, as their chips provide an optimal platform for running these workloads.
In addition to investing billions of dollars in building leading LLMs, like other companies, Jassy also emphasized Amazon’s ability to offer options to companies who want to use a foundational model in AWS and customize it for their own proprietary data, needs, and customer experience. Jassy highlighted that many companies want to do this in a way that protects their unique intellectual property (IP) from being leaked to the broader generalized model.
Amazon’s approach allows companies to leverage the power of a foundational model while still being able to tailor it to their specific needs, providing a flexible and customizable solution. This approach is particularly appealing for companies with unique data or customer experiences that cannot be easily replicated by a generalized model. “That’s what Bedrock is, which we just announced a week ago or so,” he said. It’s the company’s new API for AWS that competes directly with OpenAI’s ChatGPT and DALL-E 2 and focuses on building and scaling generative AI apps that can write, code chatbots, summarize text, classify images, and more based on text prompts.