News Hub

Amazon assigns team to train major AI model codenamed ‘Olympus’

Written by Thu 9 Nov 2023

Amazon has assembled a team to train an AI model codenamed ‘Olympus’, aiming to compete with models from OpenAI and Google.

Olympus has two trillion parameters, which could position it as one of the largest models currently under development, according to sources cited by Reuters. OpenAI’s GPT-4 model is reported to feature one trillion parameters.

More parameters in a language model result in better language understanding and text generation. However, training models with a higher number of parameters requires significant computational resources and data. This makes the AI models larger and more resource-intensive. 

Olympus is anticipated to enhance the online retail platform, improve features in the Alexa voice assistant, and provide support for Amazon Web Services (AWS). A release date for Olympus has not yet been confirmed.

The Olympus team is spearheaded by former Head of Alexa, Rohit Prasad, who will report directly to CEO Andy Jassy. As Amazon’s Head Scientist of Artificial General Intelligence, Prasad enlisted researchers from the Alexa AI and Amazon science teams to collaborate on training models.

Amazon believed that using in-house models could enhance the appeal of AWS to enterprise clients seeking top-performing models. The new model is expected to surpass Titan – a group of large language models offered to AWS cloud clients.

Amazon’s executives announced their plans to boost investments in Large Language Models and Generative AI during an earnings call in April. They also revealed their intention to reduce expenditures in their retail business, particularly in the areas of fulfilment and transportation.

In September, Amazon was set to invest £3.2 billion ($4 billion) in ChatGPT rival Anthropic. The collaboration between the two companies will make Anthropic’s foundation models and generative AI widely accessible to AWS customers.

Image Credit: Reuters

Hungry for more tech news?

Sign up for your weekly tech briefings!

Written by Thu 9 Nov 2023

Send us a correction Send us a news tip