Amazon's upcoming large language model, Olympus, could take over functionality for Alexa, Echo, and Amazon's online store.
According to an anonymous source from The Information, Amazon and its cloud unit AWS are developing the large language model "Olympus". Amazon wants to catch up with OpenAI and especially Microsoft with this model.
Microsoft in particular is using OpenAI's AI models exclusively and cleverly to drive its cloud growth, currently the most prestigious and lucrative growth business for Big Tech.
While Microsoft's cloud grew last quarter, Google's and Amazon's shrank. According to Microsoft, OpenAI's generative AI is a driving force behind its cloud growth.
Amazon uses language models in its own products for things like review summaries on product pages, offers model tuning to Bedrock companies, and is testing an LLM version of Alexa.
Amazon Olympus announcement before the end of December
Olympus could be announced by AWS as early as December, reports The Information. It is still unclear when the model will be released.
The new model is expected to outperform Amazon's current AI model, Titan, which lags significantly behind models from OpenAI, Google, and open-source models.
Amazon is also investing up to four billion US dollars in OpenAI competitor Anthropic, which launched Claude 2, a chatbot with capabilities between GPT-3.5 and GPT-4.
Amazon's problem: Google has also invested two billion US dollars in Anthropic, which is certainly a strategic move. An exclusive deal between Amazon and Anthropic, as between Microsoft and OpenAI, where Anthropic's models run exclusively in the AWS cloud, is now likely off the table.
Microsoft's exclusive OpenAI offering in the cloud currently dwarfs all other generative AI offerings, especially in Europe. In addition to Anthropic's Claude 2, Google will soon offer its exclusive multimodal LLM Gemini, which is expected to be at least at GPT-4 level.
The only company without a high-performance exclusive LLM in its cloud is Amazon. In the highly competitive cloud market, a broader and exclusive generative AI offering could be crucial, especially if the exclusive models prove to be market-leading, as is the case with GPT-4.
And things are moving fast: a first GPT-5 rumor points to an early 2024 launch and the ability to generate video in addition to text and images. But according to OpenAI CEO Sam Altman, GPT-5 is still a long way off. As with GPT-3 to GPT-3.5, there may be an intermediate step to GPT-4.5.