Amazon dedicates team to train ambitious AI model codenamed ‘Olympus’ -sources

Amazon dedicates team to train ambitious AI model codenamed ‘Olympus’ -sources

By Krystal Hu

(Reuters) – Amazon is investing millions in training an ambitious large language model (LLMs), hoping it could rival top models from OpenAI and Alphabet, two people familiar with the matter told Reuters.

The model, codenamed as “Olympus”, has 2 trillion parameters, the people said, which could make it one of the largest models being trained. OpenAI’s GPT-4 model, one of the best models available, is reported to have one trillion parameters.

The people spoke on condition of anonymity because the details of the project were not yet public.

Amazon declined to comment. The Information reported on the project name on Tuesday.

The team is spearheaded by Rohit Prasad, 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 already trained smaller models such as Titan. It has also partnered with AI model startups such as Anthropic and AI21 Labs, offering them to Amazon Web Services (AWS) users.

Amazon believes having homegrown models could make its offerings more attractive on AWS, where enterprise clients want to access top-performing models, the people familiar with the matter said, adding there is no specific timeline for releasing the new model.

LLMs are the underlying technology for AI tools that learn from huge datasets to generate human-like responses.

Training bigger AI models is more expensive given the amount of computing power required. In an earnings call in April, Amazon executives said the company would increase investment in LLMs and generative AI while cutting back on fulfillment and transportation in its retail business.

(This story has been refiled to spell out AGI in paragraph 5)

(Reporting by Krystal Hu in San Francisco. Editing by Gerry Doyle)