Amazon has dedicated a team to training the ambitious AI model codenamed “Olympus” -sources

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

Amazon ( AMZN.O ) is investing millions of dollars to train a large-scale language model (LLM) in the hope it can compete with top models from OpenAI and Alphabet ( GOOGL.O ), two people familiar with the matter told Reuters.

The model, codenamed Olympus, has 2 trillion parameters, making it one of the largest models in training, according to People. OpenAI’s GPT-4 model, which is one of the best models available, has a reported one trillion parameters.

The people spoke on condition of anonymity because details of the project had not yet been made public.

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

The team is led by Rohit Prasad, former head of Alexa, who now reports directly to CEO Andy Jassi. As chief artificial general intelligence (AGI) scientist at Amazon, Prasad brought in researchers working on Alexa’s AI and Amazon’s science team to work on training models, unifying AI efforts across the company with dedicated resources.

Amazon has already trained smaller models like the Titan. It has also partnered with AI model startups such as Anthropic and AI21 Labs and offers them to Amazon Web Services (AWS) users.

Amazon believes that having on-premise models could make its offerings on AWS more attractive, where enterprise customers want access to higher-performing models, the people familiar with the matter said, adding that there was no specific timeline for the new model’s rollout. does not have.

LLMs are the underlying technology for artificial intelligence tools that learn from massive data sets to generate human-like responses.

Due to the amount of computing power required, training larger AI models is more expensive. Amazon executives said on an earnings call in April that the company will increase investment in LLM and productive artificial intelligence while cutting back on fulfillment and shipping in its retail business.

Our Standards: The Thomson Reuters Trust Principles.