Amazon Bedrock GM Talks Generative AI Model Choice, Titan LLM, and Customer Obsession
AWS is providing an alternative to OpenAI hegemony
Why we are investing in Amazon 1p (first party) models: there are at least two core reasons that come to mind. The first one is [at] Amazon and AWS. We're very customer-obsessed…If there are certain issues related to the model, for example, say, accuracy…If that happens and a customer comes to us and says, ‘Hey. I've been really trying to deploy this customer service chatbot [and] I keep running into these issues.’ If we did not have our own first-party model, we would have to go and depend solely on a partner. When we actually build a model from scratch, we have a chance to go roll up our sleeves and actually fix that issue ourselves, which I think is something very Amazonian…
Why do we have third-party models? Now obviously choice is a very important part of it. We don't want to restrict companies to only Titan models…There is not going to be one model that is going to be the best model for everything customers will want. [They] have different use cases, and they will want to use different models for different things.
There is no one-size-fits-all approach for all customers. So we offer different services and different capabilities for different sets of customers.
Atul Deo, general manager for Amazon Bedrock, the generative AI service from AWS, recently shared the company’s view on the industry, use cases, technology, and how Amazon is serving its customers. The quote above relates to Amazon’s strategy behind providing first-party (i.e., developed by Amazon) and third-party (i.e., developed by another company) AI foundation models.
Throughout the hour-long interview, Deo talked about his experience with machine learning in customer contact center applications, transcription services, natural language processing, and generative AI. He expands on the company’s generative AI strategy as well as the technology architectures it supports.
I learned a lot from this perspective, which is not OpenAI-centric or Google-centric and considers a different way to think about the market. Let me know what you think.