Ecommerce behemoth Amazon is pouring millions of dollars into training a new large language model (LLM) codenamed ‘Olympus’ that’s hoping to challenge the likes of OpenAI and Google, according to people close to the matter.
As the AI arms race intensifies, reports claim the LLM could have 2 trillion parameters – twice the number of OpenAI’s GPT-4 model – potentially making it much more adept at understanding context, and less likely to make errors.
While this isn’t Amazon’s first foray into generative AI – with the company previously releasing an LLM called “Titan Emdeddings” in April of this year, Olympus is likely to be its biggest release to date. Here’s everything you should know about Amazon’s pending AI model, including it’s contrasts against ChatGPT.
Amazon is Developing a New AI Model ‘Olympus’
As artificial intelligence becomes Silicon Valley’s new cash crop, some of the biggest names in tech – including Meta, Microsoft, and Google – are clashing in a race to secure generative AI dominance.
So, as the competition heats up, it’s no surprise that Amazon is looking to secure its market share too. According to sources close to the matter, Amazon is currently investing millions in training a new ambitious LLM, codenamed ‘Olympus’.
🔎 Want to browse the web privately? 🌎 Or appear as if you're in another country?
Get a huge 86% off Surfshark with this special tech.co offer.
The model may be used to power new features in its online store, Alexa voice assistant on devices like Echo, and Amazon Web Services unit, according to The Information.
The team developing Olympus is being spearheaded by Rohit Prasad – former head of Alexa – who now reports to company CEO Andy Jassy. The new AI model is still under development, but sources from the Information reveal it could be announced as soon as December.
How Might Olympus Compare Against ChatGPT?
Amazon’s new LLM aims to compete directly with leading generative chatbots, but how is Olympus likely to weigh up against OpenAI’s ChatGPT?
According to those close to the matter, Amazon’s new contender will be trained on 2 trillion parameters of data – a metric referring to the values learning algorithms can change independently as they learn.
This is almost twice as many parameters as OpenAI’s GPT-4 model – which is widely considered to be the top LLM on the market. Parameters are the backbone of AI performance, so Olympus’s extra capacity will likely make the model more capable of generating accurate, human-sounding responses than GPT-4.
The jury isn’t out yet, though. OpenAI has recently announced a brand new language model, GPT-4 Turbo, which will offer a much larger context window and a knowledge of world events up to April 2023 than its previous iteration.
No information has been announced regarding GPT-4 Turbo’s parameters yet, but with OpenAI increasing its count with each LLM upgrade, it’s competition against Olympus is likely to be a close call.
Amazon is Doubling Down on Generative AI
While Amazon’s launch of Olympus is set to be its biggest AI breakthough to date, it’s not the first time the global ecommerce site has waged in the artificial intelligence race.
On September 28, the company launched Amazon Bedrock – a fully managed service that allows businesses to pick models that reflect their needs and fine-tune them with relevant data. The service essentially acts as a one-stop shop for AI models.
Amazon also recently released Titan, a similar system that is slightly more traditional than Bedrock, but offers slightly more customization over Amazon Web Service (AWS) infrastructure.
The Seattle-based company also announced it will be investing $4 billion into AI safety and research company Anthropic, giving employees and cloud customers easy access to Anthropic technology and helping Amazon gain leverage against other AI leaders.
If recent earning reports by companies like Google and Microsoft are anything to go by Amazon’s recent investments into AI are likely to pay off. However, with leading figures in politics and tech recently agreeing over the risks the technology can pose at the UK’s AI safety summit, the wider cost of rapid AI development shouldn’t be forgotten.