Hey Everyone,
Titans, Bedrocks and Pantheons of Amazon’s LLM Ecosystem
What are LLMs according to Amazon? They have something big in the works, with a focus on big.
As you know we are waiting for Google’s Gemini LLM to be announced soon. But there is another LLM that we should be aware of. Amazon was rather slow compared to its Cloud peers to invest in Generative AI products directly.
Watn access to more deep dives, you can do so here.
After its first batch of models, called Titan, got off to a rough start reports The Information, Amazon is developing a new model, codenamed Olympus various media outlets have reported this week.
Amazon claims that large language model AI will reportedly be one of the largest and most powerful systems in operation. The rumor that caught my attention is that it’s reportedly twice the size of OpenAI’s GPT-4.
The retailer has already publicly committed to invest up to $4 billion in AI startup Anthropic, and AWS has placed its bets now in Generative A.I. to some extent, though perhaps not as well as Nvidia has.
Olympus will support 2 trillion parameters, a measure of an AI model’s complexity, making it more advanced than OpenAI’s GPT-4 second most powerful model which has 1 trillion parameters, according to the sources.
Amazon needs to upgrade its Alexa devices fast, in order to keep up with Google and OpenAI. Apple have been working on their own Apple GPT. We don’t know a lot about Google Gemini except that it’s likely just weeks away from being announced in more detail.
Amazon’s AWS need a strong LLM to compete with the momentum A.I. innovation has given Microsoft’s Azure in now growing faster in the Cloud, as Azure begins to take marketshare from the once dominant AWS. Microsoft’s massive bet on the Cloud is a bit like Nvidia’s massive bet on A.I. around in the 2015-2018 time.
These foundational technologies stimulate greater advancements in the future. As A.I. has taken center stage in 2023, Nvidia and Microsoft have reaped the rewards of their dominant positioning, and Microsoft’s 13 Billion bet on OpenAI has helped a lot. They are unlikely however to keep that advantage, since Google and Amazon will pour their own Billions into catching up.
Google Gemini is being promoted as a significant advancement in natural language processing, among other things. Amazon needs Olympus to be more advanced than GPT-4 if it’s to really move the needle. With China firing on all cylinders in 2024, it will be a war of Titans around A.I. and what it can do for their services, products and enterprise customers. By all appearances, LLM dominance will mean the biggest companies become even bigger.
Microsoft, has recently reached an all-time high in its stock valuation. But in this movement so far Apple and Amazon have been consoplicyly two steps behind. This means they will be forced to make even bigger bets and spend likely each in the area of $5 Billion in 2024 to make up for their lack of agility in the space.
Amazon is doubling down on AI, investing $4 billion in Anthropic and unveiling an AI-powered Alexa. But it needs to do even more to be considered a leader or an AI innovator.
It’s not clear if Google Gemini or Amazon Olympus will be unveiled first, but they could both be announced in December 2023 or early 2024. To make matters even worse, Amazon has been fumbling a bit. Amazon reportedly delayed the launch of another AI model Titan, last year due to technical issues and the launch of ChatGPT, which executives felt was far superior to Amazon’s offering.
A world of copilots and AI assistants everywhere is now in 2023 not so far fetched as OpenAI even has unveiled a GPT store, apparently for agents that mostly function as niche chatbots. I believe Amazon and Google are likely to launch AI agent stores as well to compete directly with OpenAI on this lucrative offering.
Given Microsoft’s huge investment in OpenAI, they are likely very involved in nearly everything OpenAI does. There is a sense that even Google, Apple and Amazon have been “slow” in comparison. However Generative A.I. has just had a lot of opportunities in 2023 and first-to market races have certainly been formative.
The stock market has been juiced by this AI enthusiasm and of course it isn’t sustainable in its vibe or hype factor. The NASDAQ 100 up 41% isn’t exactly normal and neither has the exaggerated hype around Generative A.I. been. But it is good for business none the less in times where it has seemed like recession was just around the corner for nearly the last 18 months.
Amazon Olympus comes down to us from the Corporate heavens at a frantic time for the magnificent seven lordering over the world of A.I. In this light it seems like the decline of Silicon Valley may have been overstated. Certainly Nvidia with its planning and investment has been shockingly efficient in 2023 with demand for its A.I. chips at record levels. It’s stock for example, is up 225% in 2023 alone.
Amazon has gotten so big its growth could not have stayed the same. Amazon’s revenue rose 13% to $143.1 billion in the third quarter. Amazon reported an operating margin of 7.8%, the highest since it reached a record of 8.2% in the first quarter of 2021. Amazon is adopted robotics into its warehouses and company at an unprecedented rate, including a range of humanoid looking ones.
As for Olympus we’ll know more details soon as they are released. While OpenAI and new startups have been able to attract talent to their startups, it’s not clear if this has depleted the competitiveness of companies like Apple, Amazon and Meta to attract AI talent. There certainly has been a crazy exodus from Google and Meta during the last few years, while Microsoft Research has remained strong.
The Olympus team is spearheaded by Rohit Prasad, according to Reuters, who is former head of Alexa, who now reports directly to CEO Andy Jassy. As head scientist of artificial general intelligence (AGI) at Amazon, Prasad brought in researchers who had been working on Alexa AI and the Amazon science team to work on training models, uniting AI efforts across the company with dedicated resources. Amazon has realistically wasted a lot of money and effort on Alexa in the last decade and it’s been a drain on Amazon’s R&D. Now they have an even more difficult task to remain relevant in a world that’s caught somewhere between GPT-4 and GPT-5.
What can as expect of Olympus as Google Gemini might take most of the attention in the next few months? Hopefully it has some defining feature, we know that Gemini will be multi-modal in ways we’ve likely never seen before. Even GPT-4 Turbo does not feel natively multi-modal, but more like a patchwork of different things.
Project Nile
Amazon has also been developing a “conversational shopping agent” known as “Project Nile” (Insider) for its online store, and announced in September that it would integrate generative AI into its Alexa smart speaker. Given the sad history of Alexa, I will believe it when I see it.
The failed promise of Alexa to consumers was one of the worst brand hurting event that Amazon has ever undertaken. It remains to be seen if Generative A.I. can solve all the issues with it.
Read More in AI Supremacy
