According to media reports citing two insiders, Amazon is investing heavily in training a powerful Large Language Model (LLM), hoping to compete with the top models of OpenAI and Alphabet's Google.
Large language models are the fundamental technology of artificial intelligence tools that require learning and training from massive datasets to generate responses similar to those of humans.
Considering the required computing power, the cost of training larger artificial intelligence models is higher. As early as the April earnings conference call, Amazon executives had already stated that the company would increase its investment in large language models and generative artificial intelligence.
According to insiders this time, the model codenamed "Olympus" has two trillion parameters, making it one of the most powerful models being trained.
At present, OpenAI's GPT-4 model is one of the best models, and it only has one trillion parameters.
According to reports, the team is led by Rohit Prasad, former head of Amazon's voice assistant Alexa, who is now directly under CEO Andy Jesse.
Meanwhile, as the chief scientist of Amazon General Artificial Intelligence (AGI), Prasad has invited Alexa AI researchers and Amazon scientific team researchers to jointly develop the "Olympus" model.
In this year's artificial intelligence competition, Amazon is not in the first tier. Although Amazon has already trained smaller models such as Titan; And it also collaborates with artificial intelligence model startups such as Antiopic and AI21 Labs, and provides these models to Amazon Web Services (AWS).
However, the lack of its own model has always been a problem for Amazon. Insiders say that Amazon believes having its own model can make its products more attractive on AWS, as corporate customers always want to access the best performing model.
At present, the company has not officially released a specific timeline for the new model.