Amazon dedicates workforce to coach bold AI mannequin codenamed ‘Olympus’ -sources

Nov 7 (Reuters) – (This story has been refiled to spell out AGI in paragraph 5)

Amazon (AMZN.O) is investing tens of millions in coaching an bold massive language mannequin (LLMs), hoping it may rival prime fashions from OpenAI and Alphabet (GOOGL.O), two individuals conversant in the matter advised Reuters.

The mannequin, codenamed as “Olympus”, has 2 trillion parameters, the individuals stated, which may make it one of many largest fashions being educated. OpenAI’s GPT-4 mannequin, probably the greatest fashions obtainable, is reported to have one trillion parameters.

The individuals spoke on situation of anonymity as a result of the small print of the venture weren’t but public.

Amazon declined to remark. The Info reported on the venture title on Tuesday.

The workforce is spearheaded by Rohit Prasad, former head of Alexa, who now stories on to CEO Andy Jassy. As head scientist of synthetic common intelligence (AGI) at Amazon, Prasad introduced in researchers who had been engaged on Alexa AI and the Amazon science workforce to work on coaching fashions, uniting AI efforts throughout the corporate with devoted assets.

Amazon has already educated smaller fashions comparable to Titan. It has additionally partnered with AI mannequin startups comparable to Anthropic and AI21 Labs, providing them to Amazon Net Companies (AWS) customers.

Amazon believes having homegrown fashions may make its choices extra enticing on AWS, the place enterprise shoppers need to entry top-performing fashions, the individuals conversant in the matter stated, including there is no such thing as a particular timeline for releasing the brand new mannequin.

LLMs are the underlying know-how for AI instruments that be taught from enormous datasets to generate human-like responses.

Coaching greater AI fashions is dearer given the quantity of computing energy required. In an earnings name in April, Amazon executives stated the corporate would improve funding in LLMs and generative AI whereas chopping again on success and transportation in its retail enterprise.

Our Requirements: The Thomson Reuters Trust Principles.